Published as a conference paper at ICLR 2019

T HE N EURO -S YMBOLIC C ONCEPT L EARNER :

I NTERPRETING S CENES , W ORDS , AND S ENTENCES
FROM NATURAL S UPERVISION

Jiayuan Mao Chuang Gan

MIT CSAIL and IIIS, Tsinghua University MIT-IBM Watson AI Lab
mjy14@[Link] ganchuang@[Link]

Pushmeet Kohli Joshua B. Tenenbaum Jiajun Wu

arXiv:1904.12584v1 [[Link]] 26 Apr 2019

Deepmind MIT BCS, CBMM, CSAIL MIT CSAIL

pushmeet@[Link] jbt@[Link] jiajunwu@[Link]

A BSTRACT

We propose the Neuro-Symbolic Concept Learner (NS-CL), a model that learns

visual concepts, words, and semantic parsing of sentences without explicit supervi-
sion on any of them; instead, our model learns by simply looking at images and
reading paired questions and answers. Our model builds an object-based scene
representation and translates sentences into executable, symbolic programs. To
bridge the learning of two modules, we use a neuro-symbolic reasoning module
that executes these programs on the latent scene representation. Analogical to
human concept learning, the perception module learns visual concepts based on
the language description of the object being referred to. Meanwhile, the learned
visual concepts facilitate learning new words and parsing new sentences. We use
curriculum learning to guide the searching over the large compositional space
of images and language. Extensive experiments demonstrate the accuracy and
efficiency of our model on learning visual concepts, word representations, and
semantic parsing of sentences. Further, our method allows easy generalization
to new object attributes, compositions, language concepts, scenes and questions,
and even new program domains. It also empowers applications including visual
question answering and bidirectional image-text retrieval.

1 I NTRODUCTION

Humans are capable of learning visual concepts by jointly understanding vision and language (Fazly
et al., 2010; Chrupała et al., 2015; Gauthier et al., 2018). Consider the example shown in Figure 1-I.
Imagine that someone with no prior knowledge of colors is presented with the images of the red and
green cubes, paired with the questions and answers. They can easily identify the difference in objects’
visual appearance (in this case, color), and align it to the corresponding words in the questions and
answers (Red and Green). Other object attributes (e.g., shape) can be learned in a similar fashion.
Starting from there, humans are able to inductively learn the correspondence between visual concepts
and word semantics (e.g., spatial relations and referential expressions, Figure 1-II), and unravel
compositional logic from complex questions assisted by the learned visual concepts (Figure 1-III,
also see Abend et al. (2017)).
Motivated by this, we propose the neuro-symbolic concept learner (NS-CL), which jointly learns
visual perception, words, and semantic language parsing from images and question-answer pairs.
NS-CL has three modules: a neural-based perception module that extracts object-level representa-
tions from the scene, a visually-grounded semantic parser for translating questions into executable
programs, and a symbolic program executor that reads out the perceptual representation of objects,
classifies their attributes/relations, and executes the program to obtain an answer.

Project page: [Link]

1
Published as a conference paper at ICLR 2019

I. Learning basic, object-based concepts. II. Learning relational concepts based on referential expressions.
Q: How many objects are right of the red object?
Q: What’s the color of the object? A: 2.
A: Red. Q: How many objects have the same material as the cube?
Q: Is there any cube? A: 2
A: Yes.
Q: What’s the color of the object? III. Interpret complex questions from visual cues.
A: Green. Q: How many objects are both right of the green cylinder
Q: Is there any cube? and have the same material as the small blue ball?
A: Yes. A: 3

Figure 1: Humans learn visual concepts, words, and semantic parsing jointly and incrementally.
I. Learning visual concepts (red vs. green) starts from looking at simple scenes, reading simple
questions, and reasoning over contrastive examples (Fazly et al., 2010). II. Afterwards, we can
interpret referential expressions based on the learned object-based concepts, and learn relational
concepts (e.g., on the right of, the same material as). III Finally, we can interpret complex questions
from visual cues by exploiting the compositional structure.

NS-CL learns from natural supervision (i.e., images and QA pairs), requiring no annotations on
images or semantic programs for sentences. Instead, analogical to human concept learning, it learns
via curriculum learning. NS-CL starts by learning representations/concepts of individual objects from
short questions (e.g., What’s the color of the cylinder?) on simple scenes (≤3 objects). By doing
so, it learns object-based concepts such as colors and shapes. NS-CL then learns relational concepts
by leveraging these object-based concepts to interpret object referrals (e.g., Is there a box right of a
cylinder?). The model iteratively adapts to more complex scenes and highly compositional questions.
NS-CL’s modularized design enables interpretable, robust, and accurate visual reasoning: it achieves
state-of-the-art performance on the CLEVR dataset (Johnson et al., 2017a). More importantly, it
naturally learns disentangled visual and language concepts, enabling combinatorial generalization
w.r.t. both visual scenes and semantic programs. In particular, we demonstrate four forms of
generalization. First, NS-CL generalizes to scenes with more objects and longer semantic programs
than those in the training set. Second, it generalizes to new visual attribute compositions, as
demonstrated on the CLEVR-CoGenT (Johnson et al., 2017a) dataset. Third, it enables fast adaptation
to novel visual concepts, such as learning a new color. Finally, the learned visual concepts transfer to
new tasks, such as image-caption retrieval, without any extra fine-tuning.

2 R ELATED W ORK

Our model is related to research on joint learning of vision and natural language. In particular, there
are many papers that learn visual concepts from descriptive languages, such as image-captioning
or visually-grounded question-answer pairs (Kiros et al., 2014; Shi et al., 2018; Mao et al., 2016;
Vendrov et al., 2016; Ganju et al., 2017), dense language descriptions for scenes (Johnson et al.,
2016), video-captioning (Donahue et al., 2015) and video-text alignment (Zhu et al., 2015).
Visual question answering (VQA) stands out as it requires understanding both visual content and
language. The state-of-the-art approaches usually use neural attentions (Malinowski & Fritz, 2014;
Chen et al., 2015; Yang et al., 2016; Xu & Saenko, 2016). Beyond question answering, Johnson
et al. (2017a) proposed the CLEVR (VQA) dataset to diagnose reasoning models. CLEVR contains
synthetic visual scenes and questions generated from latent programs. Table 1 compares our model
with state-of-the-art visual reasoning models (Andreas et al., 2016; Suarez et al., 2018; Santoro et al.,
2017) along four directions: visual features, semantics, inference, and the requirement of extra labels.
For visual representations, Johnson et al. (2017b) encoded visual scenes into a convolutional feature
map for program operators. Mascharka et al. (2018); Hudson & Manning (2018) used attention as
intermediate representations for transparent program execution. Recently, Yi et al. (2018) explored an
interpretable, object-based visual representation for visual reasoning. It performs well, but requires
fully-annotated scenes during training. Our model also adopts an object-based visual representation,
but the representation is learned only based on natural supervision (questions and answers).
Anderson et al. (2018) also proposed to represent the image as a collection of convolutional object
features and gained substantial improvements on VQA. Their model encodes questions with neural

2
Published as a conference paper at ICLR 2019

Extra Labels
Models Visual Features Semantics Inference
# Prog. Attr.
FiLM (Perez et al., 2018) Convolutional Implicit 0 No Feature Manipulation
IEP (Johnson et al., 2017b) Convolutional Explicit 700K No Feature Manipulation
MAC (Hudson & Manning, 2018) Attentional Implicit 0 No Feature Manipulation
Stack-NMN (Hu et al., 2018) Attentional Implicit 0 No Attention Manipulation
TbD (Mascharka et al., 2018) Attentional Explicit 700K No Attention Manipulation
NS-VQA (Yi et al., 2018) Object-Based Explicit 0.2K Yes Symbolic Execution
NS-CL Object-Based Explicit 0 No Symbolic Execution

Table 1: Comparison with other frameworks on the CLEVR VQA dataset, w.r.t. visual features,
implicit or explicit semantics and supervisions.

Back-propagation
Visual Representation Concept Embeddings
1 2
Θ# Obj 1 Sphere ……
Obj 2
3 4 Obj 3 Symbolic Reasoning
Obj 4 Back-propagation Answer: Cylinder
Groundtruth: Box
Semantic Parsing (Candidate Interpretations)
Θ"
✓ Query(Shape, Filter(Red, Relate(Left, Filter(Sphere))))
Q: What is the shape of
☓ Query(Shape, Filter(Sphere, Relate(Left, Filter(Red))))
the red object left of the ☓ Exist(AERelate(Shape, Filter(Red, Relate(Left, Filter(Sphere))))) REINFORCE
sphere? ……

Figure 2: We propose to use neural symbolic reasoning as a bridge to jointly learn visual concepts,
words, and semantic parsing of sentences.

networks and answers the questions by question-conditioned attention over the object features. In
contrast, NS-CL parses question inputs into programs and executes them on object features to get the
answer. This makes the reasoning process interpretable and supports combinatorial generalization
over quantities (e.g., counting objects). Our model also learns general visual concepts and their
association with symbolic representations of language. These learned concepts can then be explicitly
interpreted and deployed in other vision-language applications such as image caption retrieval.
There are two types of approaches in semantic sentence parsing for visual reasoning: implicit pro-
grams as conditioned neural operations (e.g., conditioned convolution and dual attention) (Perez et al.,
2018; Hudson & Manning, 2018) and explicit programs as sequences of symbolic tokens (Andreas
et al., 2016; Johnson et al., 2017b; Mascharka et al., 2018). As a representative, Andreas et al.
(2016) build modular and structured neural architectures based on programs for answering questions.
Explicit programs gain better interpretability, but usually require extra supervision such as ground-
truth program annotations for training. This restricts their application. We propose to use visual
grounding as distant supervision to parse questions in natural languages into explicit programs, with
zero program annotations. Given the semantic parsing of questions into programs, Yi et al. (2018)
proposed a purely symbolic executor for the inference of the answer in the logic space. Compared
with theirs, we propose a quasi-symbolic executor for VQA.
Our work is also related to learning interpretable and disentangled representations for visual scenes
using neural networks. Kulkarni et al. (2015) proposed convolutional inverse graphics networks for
learning and inferring pose of faces, while Yang et al. (2015) learned disentangled representation of
pose of chairs from images. Wu et al. (2017) proposed the neural scene de-rendering framework as
an inverse process of any rendering process. Siddharth et al. (2017); Higgins et al. (2018) learned
disentangled representations using deep generative models. In contrast, we propose an alternative
representation learning approach through joint reasoning with language.

3 N EURO -S YMBOLIC CONCEPT LEARNER

We present our neuro-symbolic concept learner, which uses a symbolic reasoning process to bridge
the learning of visual concepts, words, and semantic parsing of sentences without explicit annotations

3
Published as a conference paper at ICLR 2019

Cube
Sphere
ShapeOf( ) = Cylinder
ShapeOf(Obj1) Cube
Similarity( , ) = 0.99
Visual Perception Module Visual Attribute Operators Visual-Semantic Space Concept Embeddings

Figure 3: We treat attributes such as Shape and Color as neural operators. The operators map
object representations into a visual-semantic space. We use similarity-based metric to classify objects.

for any of them. We first use a visual perception module to construct an object-based representation
for a scene, and run a semantic parsing module to translate a question into an executable program. We
then apply a quasi-symbolic program executor to infer the answer based on the scene representation.
We use paired images, questions, and answers to jointly train the visual and language modules.
Shown in Figure 2, given an input image, the visual perception module detects objects in the scene
and extracts a deep, latent representation for each of them. The semantic parsing module translates
an input question in natural language into an executable program given a domain specific language
(DSL). The generated programs have a hierarchical structure of symbolic, functional modules, each
fulfilling a specific operation over the scene representation. The explicit program semantics enjoys
compositionality, interpretability, and generalizability.
The program executor executes the program upon the derived scene representation and answers
the question. Our program executor works in a symbolic and deterministic manner. This feature
ensures a transparent execution trace of the program. Our program executor has a fully differentiable
design w.r.t. the visual representations and the concept representations, which supports gradient-based
optimization during training.

3.1 M ODEL DETAILS

Visual perception. Shown in Figure 2, given the input image, we use a pretrained Mask R-CNN
(He et al., 2017) to generate object proposals for all objects. The bounding box for each single object
paired with the original image is then sent to a ResNet-34 (He et al., 2015) to extract the region-based
(by RoI Align) and image-based features respectively. We concatenate them to represent each object.
Here, the inclusion of the representation of the full scene adds the contextual information, which is
essential for the inference of relative attributes such as size or spatial position.

Concept quantization. Visual reasoning requires determining an object’s attributes (e.g., its color
or shape). We assume each visual attribute (e.g., shape) contains a set of visual concept (e.g.,
Cube). In NS-CL, visual attributes are implemented as neural operators , mapping the object
representation into an attribute-specific embedding space. Figure 3 shows an inference an object’s
shape. Visual concepts that belong to the shape attribute, including Cube, Sphere and Cylinder,
are represented as vectors in the shape embedding space. These concept vectors are also learned
along the process. We measure the cosine distances h·, ·i between these vectors to determine
the shape of the object. Specifically,

 we compute the probability that an object oi is a cube by
σ hShapeOf(oi ), v Cube i − γ τ , where ShapeOf(·) denotes the neural operator, v Cube the concept
embedding of Cube and σ the Sigmoid function. γ and τ are scalar constants for scaling and
shifting the values of similarities. We classify relational concepts (e.g., Left) between a pair of
objects similarly, except that we concatenate the visual representations for both objects to form the
representation of their relation.

DSL and semantic parsing. The semantic parsing module translates a natural language question
into an executable program with a hierarchy of primitive operations, represented in a domain-specific
language (DSL) designed for VQA. The DSL covers a set of fundamental operations for visual
reasoning, such as filtering out objects with certain concepts or query the attribute of an object. The
operations share the same input and output interface, and thus can be compositionally combined
to form programs of any complexity. We include a complete specification of the DSL used by our
framework in the Appendix A.

4
Published as a conference paper at ICLR 2019

A. Curriculum concept learning B. Illustrative execution of NS-CL

Initialized with DSL and executor. Q: Does the red object left of the green
cube have the same shape as the
Lesson1: Object-based questions. purple matte thing?

Q: What is the shape of the red object? Step1: Visual Parsing

A: Cube. 1
Obj 1 2
Obj 2
Lesson2: Relational questions. Obj 3 3 4
Obj 4
Q: How many cubes are behind the
sphere? Step2, 3: Semantic Parsing and Program Execution
A: 3
Program Representations Concepts Outputs
Lesson3: More complex questions.
Filter Green Cube
Q: Does the red object left of the green
cube have the same shape as the
Relate Object 2
purple matte thing? Left
A: No
Filter Red
Deploy: complex scenes, complex questions
Q: Does the matte thing behind the big
Filter Purple Matte
sphere have the same color as the
cylinder left of the small matte cube?
Object 1 Object 3
A: No. AEQuery Shape No (0.98)

Figure 4: A. Demonstration of the curriculum learning of visual concepts, words, and semantic parsing
of sentences by watching images and reading paired questions and answers. Scenes and questions of
different complexities are illustrated to the learner in an incremental manner. B. Illustration of our
neuro-symbolic inference model for VQA. The perception module begins with parsing visual scenes
into object-based deep representations, while the semantic parser parse sentences into executable
programs. A symbolic execution process bridges two modules.

Our semantic parser generates the hierarchies of latent programs in a sequence to tree manner (Dong
& Lapata, 2016). We use a bidirectional GRU (Cho et al., 2014) to encode an input question, which
outputs a fixed-length embedding of the question. A decoder based on GRU cells is applied to
the embedding, and recovers the hierarchy of operations as the latent program. Some operations
takes concepts their parameters, such as Filter( Red ) and Query( Shape ). These concepts are
chosen from all concepts appeared in the input question. Figure 4(B) shows an example, while more
details can be found in Appendix B.

Quasi-symbolic program execution. Given the latent program recovered from the question in
natural language, a symbolic program executor executes the program and derives the answer based
on the object-based visual representation. Our program executor is a collection of deterministic
functional modules designed to realize all logic operations specified in the DSL. Figure 4(B) shows
an illustrative execution trace of a program.
To make the execution differentiable w.r.t. visual representations, we represent the intermediate results
in a probabilistic manner: a set of objects is represented by a vector, as the attention mask over all
objects in the scene. Each element, Maski ∈ [0, 1] denotes the probability that the i-th object of the
scene belongs to the set. For example, shown in Figure 4(B), the first Filter operation outputs
a mask of length 4 (there are in total 4 objects in the scene), with each element representing the
probability that the corresponding object is selected out (i.e., the probability that each object is a
green cube). The output “mask” on the objects will be fed into the next module (Relate in this
case) as input and the execution of programs continues. The last module outputs the final answer to
the question. We refer interested readers to Appendix C for the implementation of all operators.

3.2 T RAINING PARADIGM

Optimization objective. The optimization objective of NS-CL is composed of two parts: concept
learning and language understanding. Our goal is to find the optimal parameters Θv of the visual

5
Published as a conference paper at ICLR 2019

perception module Perception (including the ResNet-34 for extracting object features, attribute
operators. and concept embeddings) and Θs of the semantic parsing module SemanticParse, to
maximize the likelihood of answering the question Q correctly:
Θv , Θs ← arg max EP [Pr[A = Executor(Perception(S; Θv ), P )]], (1)
Θv ,Θs

where P denotes the program, A the answer, S the scene, and Executor the quasi-symbolic executor.
The expectation is taken over P ∼ SemanticParse(Q; Θs ).
Recall the program executor is fully differentiable w.r.t. the visual representation. We com-
pute the gradient w.r.t. Θv as ∇Θv EP [DKL (Executor(Perception(S; Θv ), P )kA)]. We use RE-
INFORCE (Williams, 1992) to optimize the semantic parser Θs : ∇Θs = EP [r · log Pr[P =
SemanticParse(Q; Θs )]], where the reward r = 1 if the answer is correct and 0 otherwise. We also
use off-policy search to reduce the variance of REINFORCE, the detail of which can be found in
Appendix D.

Curriculum visual concept learning. Motivated by human concept learning as in Figure 1, we

employ a curriculum learning approach to help joint optimization. We heuristically split the training
samples into four stages (Figure 4(A)): first, learning object-level visual concepts; second, learning
relational questions; third, learning more complex questions with perception modules fixed; fourth,
joint fine-tuning of all modules. We found that this is essential to the learning of our neuro-symbolic
concept learner. We include more technical details in Appendix E.

4 E XPERIMENTS

We demonstrate the following advantages of our NS-CL. First, it learns visual concepts with remark-
able accuracy; second, it allows data-efficient visual reasoning on the CLEVR dataset (Johnson et al.,
2017a); third, it generalizes well to new attributes, visual composition, and language domains.
We train NS-CL on 5K images (<10% of CLEVR’s 70K training images). We generate 20 questions
for each image for the entire curriculum learning process. The Mask R-CNN module is pretrained on
4K generated CLEVR images with bounding box annotations, following Yi et al. (2018).

4.1 V ISUAL CONCEPT LEARNING

Classification-based concept evaluation. Our model treats attributes as neural operators that map
latent object representations into an attribute-specific embedding space (Figure 3). We evaluate the
concept quantization of objects in the CLEVR validation split. Our model can achieve near perfect
classification accuracy (∼99%) for all object properties, suggesting it effectively learns generic
concept representations. The result for spatial relations is relatively lower, because CLEVR does not
have direct queries on the spatial relation between objects. Thus, spatial relation concepts can only
be learned indirectly.

Count-based concept evaluation. The SOTA methods do not provide interpretable representation
on individual objects (Johnson et al., 2017a; Hudson & Manning, 2018; Mascharka et al., 2018) .
To evaluate the visual concepts learned by such models, we generate a synthetic question set. The
diagnostic question set contains simple questions as the following form: “How many red objects are
there?”. We evaluate the performance on all concepts appeared in the CLEVR dataset.
Table 2 summarizes the results compared with strong baselines, including methods based on con-
volutional features (Johnson et al., 2017b) and those based on neural attentions (Mascharka et al.,
2018; Hudson & Manning, 2018). Our approach outperforms IEP by a significant margin (8%)
and attention-based baselines by >2%, suggesting object-based visual representations and symbolic
reasoning helps to interpret visual concepts.

4.2 DATA - EFFICIENT AND INTERPRETABLE VISUAL REASONING

NS-CL jointly learns visual concepts, words and semantic parsing by watching images and reading
paired questions and answers. It can be directly applied to VQA.

6
Published as a conference paper at ICLR 2019

Visual Mean Color Mat. Shape Size Accuracy Accuracy

Model Visual
(100% Data) (10% Data)
IEP Conv. 90.6 91.0 90.0 89.9 90.6
TbD Attn. 99.1 54.2
MAC Attn. 95.9 98.0 91.4 94.4 94.2 TbD-Object Obj. 84.1 52.6
TbD (hres.) Attn. 96.5 96.6 92.2 95.4 92.6 TbD-Mask Attn. 99.0 55.0
NS-CL Obj. 98.7 99.0 98.7 98.1 99.1 MAC Attn. 98.9 67.3
MAC-Object Obj. 79.5 51.2
MAC-Mask Attn. 98.7 68.4
Table 2: We also evaluate the learned visual con- NS-CL Obj. 99.2 98.9
cepts using a diagnostic question set containing
simple questions such as “How many red ob- Table 3: We compare different variants of base-
jects are there?”. NS-CL outperforms both con- lines for a systematic study on visual features and
volutional and attentional baselines. The sug- data efficiency. Using only 10% of the training
gested object-based visual representation and images, our model is able to achieve a compara-
symbolic reasoning approach perceives better ble results with the baselines trained on the full
interpretation of visual concepts. dataset. See the text for details.

Table 4 summarizes results on the CLEVR validation split. Our model achieves the state-of-the-
art performance among all baselines using zero program annotations, including MAC (Hudson &
Manning, 2018) and FiLM (Perez et al., 2018). Our model achieves comparable performance with
the strong baseline TbD-Nets (Mascharka et al., 2018), whose semantic parser is trained using 700K
programs in CLEVR (ours need 0). The recent NS-VQA model from Yi et al. (2018) achieves better
performance on CLEVR; however, their system requires annotated visual attributes and program
traces during training, while our NS-CL needs no extra labels.
Here, the visual perception module is pre-trained on ImageNet (Deng et al., 2009). Without pre-
training, the concept learning accuracies drop by 0.2% on average and the QA accuracy drops by
0.5%. Meanwhile, NS-CL recovers the underlying programs of questions accurately (> 99.9%
accuracy). NS-CL can also detect ambiguous or invalid programs and indicate exceptions. Please see
Appendix F for more details. NS-CL can also be applied to other visual reasoning testbeds. Please
refer to Appendix G.1 for our results on the Minecraft dataset (Yi et al., 2018).
For a systematic study on visual features and data efficiency, we implement two variants of the
baseline models: TbD-Object and MAC-Object. Inspired by (Anderson et al., 2018), instead of the
input image, TbD-Object and MAC-Object take a stack of object features as input. TbD-Mask and
MAC-Mask integrate the masks of objects by using them to guide the attention over the images.
Table 3 summarizes the results. Our model outperforms all baselines on data efficiency. This comes
from the full disentanglement of visual concept learning and symbolic reasoning: how to execute
program instructions based on the learned concepts is programmed. TbD-Object and MAC-Object
demonstrate inferior results in our experiments. We attribute this to the design of model architectures
and have a detailed analysis in Appendix F.3. Although TbD-Mask and MAC-Mask do not perform
better than the originals, we find that using masks to guide attentions speeds up the training.
Besides achieving a competitive performance on the visual reasoning testbeds, by leveraging both
object-based representation and symbolic reasoning, out model learns fully interpretable visual
concepts: see Appendix H for qualitative results on various datasets.

4.3 G ENERALIZATION TO NEW ATTRIBUTES AND COMPOSITIONS

Generalizing to new visual compositions. The CLEVR-CoGenT dataset is designed to evaluate

models’ ability to generalize to new visual compositions. It has two splits: Split A only contains
gray, blue, brown and yellow cubes, but red, green, purple, and cyan cylinders; split B imposes the
opposite color constraints on cubes and cylinders. If we directly learn visual concepts on split A, it
overfits to classify shapes based on the color, leading to a poor generalization to split B.
Our solution is based on the idea of seeing attributes as operators. Specifically, we jointly train the
concept embeddings (e.g., Red, Cube, etc.) as well as the semantic parser on split A, keeping pre-
trained, frozen attribute operators. As we learn distinct representation spaces for different attributes,
our model achieves an accuracy of 98.8% on split A and 98.9% on split B.

7
Published as a conference paper at ICLR 2019

Test
Prog. Cmp. Query Cmp. Model
Model Overall Count Exist Split A Split B Split C Split D
Anno. Num. Attr. Attr.

Human N/A 92.6 86.7 86.4 96.6 95.0 96.0 MAC 97.3 N/A 92.9 N/A
IEP 96.1 92.1 91.5 90.9
NMN 700K 72.1 52.5 72.7 79.3 79.0 78.0 TbD 98.8 94.5 94.3 91.9
N2NMN 700K 88.8 68.5 84.9 85.7 90.0 88.8
IEP 700K 96.9 92.7 98.7 97.1 98.1 98.9 NS-CL 98.9 98.9 98.7 98.8
DDRprog 700K 98.3 96.5 98.4 98.8 99.1 99.0
TbD 700K 99.1 97.6 99.4 99.2 99.5 99.6 Figure 5: We test the combinatorial gen-
RN 0 95.5 90.1 93.6 97.8 97.1 97.9
eralization w.r.t. the number of objects
FiLM 0 97.6 94.5 93.8 99.2 99.2 99.0 in scenes and the complexity of ques-
MAC 0 98.9 97.2 99.4 99.5 99.3 99.5 tions (i.e. the depth of the program trees).
We makes four split of the data contain-
NS-CL 0 98.9 98.2 99.0 98.8 99.3 99.1 ing various complexities of scenes and
questions. Our object-based visual repre-
Table 4: Our model outperforms all baselines using no sentation and explicit program semantics
program annotations. It achieves comparable results enjoys the best (and almost-perfect) com-
with models trained by full program annotations such binatorial generalization compared with
as TbD. strong baselines.

Split A Split B Split C Split D

Q: What’s the shape of the big Q: What size is the cylinder that Q: What’s the shape of the big Q: What size is the cylinder that
yellow thing? is left of the cyan thing that is in yellow thing? is left of the cyan thing that is in
front of the big sphere? front of the gray cube?

Figure 6: Samples collected from four splits in Section 4.3 for illustration. Models are trained on
split A but evaluated on all splits for testing the combinatorial generalization.

Generalizing to new visual concepts. We expect the process of concept learning can take place
in an incremental manner: having learned 7 different colors, humans can learn the 8-th color
incrementally and efficiently. To this end, we build a synthetic split of the CLEVR dataset to replicate
the setting of incremental concept learning. Split A contains only images without any purple objects,
while split B contains images with at least one purple object. We train all the models on split A
first, and finetune them on 100 images from split B. We report the final QA performance on split B’s
validation set. All models use a pre-trained semantic parser on the full CLEVR dataset.
Our model performs a 93.9% accuracy on the QA test in Split B, outperforming the convolutional
baseline IEP (Johnson et al., 2017b) and the attentional baseline TbD (Mascharka et al., 2018) by
4.6% and 6.1% respectively. The acquisition of Color operator brings more efficient learning of
new visual concepts.

4.4 C OMBINATORIAL GENERALIZATION TO NEW SCENES AND QUESTIONS

Having learned visual concepts on small-scale scenes (containing only few objects) and simple
questions (only single-hop questions), we humans can easily generalize the knowledge to larger-scale
scenes and to answer complex questions. To evaluate this, we split the CLEVR dataset into four parts:
Split A contains only scenes with less than 6 objects, and questions whose latent programs having a
depth less than 5; Split B contains scenes with less than 6 objects, but arbitrary questions; Split C
contains arbitrary scenes, but restricts the program depth being less than 5; Split D contains arbitrary
scenes and questions. Figure 6 shows some illustrative samples.
As VQA baselines are unable to count a set of objects of arbitrary size, for a fair comparison,
all programs containing the “count” operation over > 6 objects are removed from the set. For

8
Published as a conference paper at ICLR 2019

Model Retrieval Accuracy Model Retrieval Accuracy

IEP 95.5 CNN-LSTM 68.9
TbD 97.0 NS-CL 97.0
NS-CL 96.9
Caption: There is a big yellow
cylinder in front of a gray object.
(c) Image-caption retrieval accuracy
(b) Image-caption retrieval accuracy on the full dataset. Our model outper-
(a) An illustrative pair of im- on a subset of data. Our model forms baselines and requires no extra
age and caption in our syn- archives comparable results with training or fine-tuning of the visual
thetic dataset. VQA baselines. perception module.

Table 5: We introduce a new simple DSL for image-caption retrieval to evaluate how well the learned
visual concepts transfer. Due to the difference between VQA and caption retrieval, VQA baselines
are only able to infer the result on a partial set of data. The learned object-based visual concepts can
be directly transferred into the new domain for free.

methods using explicit program semantics, the semantic parser is pre-trained on the full dataset and
fixed. Methods with implicit program semantics (Hudson & Manning, 2018) learn an entangled
representation for perception and reasoning, and cannot trivially generalize to more complex programs.
We only use the training data from the Split A and then quantify the generalization ability on other
three splits. Shown in Table 5, our NS-CL leads to almost-perfect generalization to larger scenes and
more complex questions, outperforming all baselines by at least 4% in QA accuracy.

4.5 E XTENDING TO OTHER PROGRAM DOMAIN

The learned visual concepts can also be used in other domains such as image retrieval. With the
visual scenes fixed, the learned visual concepts can be directly transferred into the new domain. We
only need to learn the semantic parsing of natural language into the new DSL.
We build a synthetic dataset for image retrieval and adopt a DSL from scene graph–based image
retrieval (Johnson et al., 2015). The dataset contains only simple captions: “There is an <object A>
<relation> <object B>.” (e.g., There is a box right of a cylinder). The semantic parser learns to
extract corresponding visual concepts (e.g., box, right, and cylinder) from the sentence. The
program can then be executed on the visual representation to determine if the visual scene contains
such relational triples.
For simplicity, we treat retrieval as classifying whether a relational triple exists in the image. This
functionality cannot be directly implemented on the CLEVR VQA program domain, because ques-
tions such as “Is there a box right of a cylinder” can be ambiguous if there exist multiple cylinders
in the scene. Due to the entanglement of the visual representation with the specific DSL, baselines
trained on CLEVR QA can not be directly applied to this task. For a fair comparison with them, we
show the result in Table 5b on a subset of the generated image-caption pairs where the underlying
programs have no ambiguity regarding the reference of object B. A separate semantic parser is
trained for the VQA baselines, which translates captions into a CLEVR QA-compatible program
(e.g., Exist(Filter(Box, Relate(Right, Filter(Cylinder))).
Table 5c compares our NS-CL against typical image-text retrieval baselines on the full image-caption
dataset. Without any annotations of the sentence semantics, our model learns to parse the captions
into the programs in the new DSL. It outperforms the CNN-LSTM baseline by 30%.

4.6 E XTENDING TO NATURAL IMAGES AND LANGUAGE

We further conduct experiments on MS-COCO (Lin et al., 2014) images. Results are presented on
the VQS dataset (Gan et al., 2017). VQS contains a subset of images and questions from the original
VQA 1.0 dataset (Antol et al., 2015). All questions in the VQS dataset can be visually grounded: each
question is associated with multiple image regions, annotated by humans as essential for answering
the question. Figure 7 illustrates an execution trace of NS-CL on VQS.
We use a syntactic dependency parser to extract programs and concepts from language (Andreas et al.,
2016; Schuster et al., 2015). The object proposals and features are extracted from models pre-trained
on the MS-COCO dataset and the ImageNet dataset, respectively. Illustrated in Figure 7, our model

9
Published as a conference paper at ICLR 2019

Filter:
Fire_Hydrant

Yellow (0.87) ✓ Model Accuracy

MLP 43.9
Program MAC 46.2
Q: What color is the NS-CL 44.3
fire hydrant? Filter Query
Fire_Hydrant Color

Figure 7: Left: An example image-question pair from the VQS dataset and the corresponding
execution trace of NS-CL. Right: Results on the VQS test set. Our model achieves a comparable
results with the baselines.

Concept: Horse Concept: Person On a Skateboard

Figure 8: Concepts learned from VQS, including object categories, attributes, and relations.

shows competitive performance on QA accuracy, comparable with the MLP baseline (Jabri et al.,
2016) and the MAC network (Hudson & Manning, 2018). Additional illustrative execution traces of
NS-CL are in Appendix H. Beyond answering questions, NS-CL effectively learns visual concepts
from data. Figure 8 shows examples of the learned visual concepts, including object categories,
attributes, and relations. Experiment setup and implementation details are in Appendix G.2.
In this paper, we focus on a neuro-symbolic framework that learns visual concepts about object
properties and relations. Indeed, visual question answering requires AI systems to reason about more
general concepts such as events or activities (Levin, 1993). We leave the extension of NS-CL along
this direction and its application to general VQA datasets (Antol et al., 2015) as future work.

5 D ISCUSSION AND F UTURE W ORK

We presented a method that jointly learns visual concepts, words, and semantic parsing of sentences
from natural supervision. The proposed framework, NS-CL, learns by looking at images and reading
paired questions and answers, without any explicit supervision such as class labels for objects. Our
model learns visual concepts with remarkable accuracy. Based upon the learned concepts, our model
achieves good results on question answering, and more importantly, generalizes well to new visual
compositions, new visual concepts, and new domain specific languages.
The design of NS-CL suggests multiple research directions. First, constructing 3D object-based
representations for realistic scenes needs further exploration (Anderson et al., 2018; Baradel et al.,
2018). Second, our model assumes a domain-specific language for describing formal semantics. The
integration of formal semantics into the processing of complex natural language would be meaningful
future work (Artzi & Zettlemoyer, 2013; Oh et al., 2017). We hope our paper could motivate future
research in visual concept learning, language learning, and compositionality.
Our framework can also be extended to other domains such as video understanding and robotic
manipulation. Here, we would need to discover semantic representations for actions and interactions
(e.g., push) beyond static spatial relations. Along this direction, researchers have studied building
symbolic representations for skills (Konidaris et al., 2018) and learning instruction semantics from
interaction (Oh et al., 2017) in constrained setups. Applying neuro-symbolic learning frameworks for
concepts and skills would be meaningful future work toward robotic learning in complex interactive
environments.

10
Published as a conference paper at ICLR 2019

Acknowledgements. We thank Kexin Yi, Haoyue Shi, and Jon Gauthier for helpful discussions
and suggestions. This work was supported in part by the Center for Brains, Minds and Machines
(NSF STC award CCF-1231216), ONR MURI N00014-16-1-2007, MIT-IBM Watson AI Lab, and
Facebook.

R EFERENCES

Omri Abend, Tom Kwiatkowski, Nathaniel J Smith, Sharon Goldwater, and Mark Steedman. Boot-
strapping language acquisition. Cognition, 2017.

Peter Anderson, Xiaodong He, Chris Buehler, Damien Teney, Mark Johnson, Stephen Gould, and Lei
Zhang. Bottom-up and top-down attention for image captioning and visual question answering. In
CVPR, 2018.

Jacob Andreas, Marcus Rohrbach, Trevor Darrell, and Dan Klein. Learning to compose neural
networks for question answering. In NAACL-HLT, 2016.

Stanislaw Antol, Aishwarya Agrawal, Jiasen Lu, Margaret Mitchell, Dhruv Batra, C Lawrence Zitnick,
and Devi Parikh. VQA: Visual question answering. In ICCV, 2015.

Yoav Artzi and Luke Zettlemoyer. Weakly supervised learning of semantic parsers for mapping
instructions to actions. TACL, 2013.

Fabien Baradel, Natalia Neverova, Christian Wolf, Julien Mille, and Greg Mori. Object level visual
reasoning in videos. In ECCV, 2018.

Kan Chen, Jiang Wang, Liang-Chieh Chen, Haoyuan Gao, Wei Xu, and Ram Nevatia. Abc-cnn: An
attention based convolutional neural network for visual question answering. arXiv:1511.05960,
2015.

Kyunghyun Cho, Bart Van Merriënboer, Caglar Gulcehre, Dzmitry Bahdanau, Fethi Bougares, Holger
Schwenk, and Yoshua Bengio. Learning phrase representations using rnn encoder-decoder for
statistical machine translation. In EMNLP, 2014.

Grzegorz Chrupała, Akos Kádár, and Afra Alishahi. Learning language through pictures. In ACL,
2015.

Jia Deng, Wei Dong, Richard Socher, Li-Jia Li, Kai Li, and Li Fei-Fei. Imagenet: A large-scale
hierarchical image database. In CVPR, 2009.

Jeffrey Donahue, Lisa Anne Hendricks, Sergio Guadarrama, Marcus Rohrbach, Subhashini Venu-
gopalan, Kate Saenko, and Trevor Darrell. Long-term recurrent convolutional networks for visual
recognition and description. In CVPR, 2015.

Li Dong and Mirella Lapata. Language to logical form with neural attention. In ACL, 2016.

Afsaneh Fazly, Afra Alishahi, and Suzanne Stevenson. A probabilistic computational model of
cross-situational word learning. Cognit. Sci., 2010.

Chuang Gan, Yandong Li, Haoxiang Li, Chen Sun, and Boqing Gong. VQS: Linking segmentations to
questions and answers for supervised attention in vqa and question-focused semantic segmentation.
In ICCV, 2017.

Siddha Ganju, Olga Russakovsky, and Abhinav Gupta. What’s in a question: Using visual questions
as a form of supervision. In CVPR, 2017.

Jon Gauthier, Roger Levy, and Joshua B Tenenbaum. Word learning and the acquisition of syntactic–
semantic overhypotheses. In CogSci, 2018.

Jonas Gehring, Michael Auli, David Grangier, Denis Yarats, and Yann N Dauphin. Convolutional
sequence to sequence learning. In ICML, 2017.

11
Published as a conference paper at ICLR 2019

Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image
recognition. In CVPR, 2015.
Kaiming He, Georgia Gkioxari, Piotr Dollár, and Ross Girshick. Mask R-CNN. In ICCV, 2017.
Irina Higgins, Nicolas Sonnerat, Loic Matthey, Arka Pal, Christopher P Burgess, Matthew Botvinick,
Demis Hassabis, and Alexander Lerchner. Scan: learning abstract hierarchical compositional
visual concepts. In ICLR, 2018.
Ronghang Hu, Jacob Andreas, Trevor Darrell, and Kate Saenko. Explainable neural computation via
stack neural module networks. In ECCV, 2018.
Drew A Hudson and Christopher D Manning. Compositional attention networks for machine
reasoning. In ICLR, 2018.
Allan Jabri, Armand Joulin, and Laurens van der Maaten. Revisiting visual question answering
baselines. In ECCV, 2016.
Justin Johnson, Ranjay Krishna, Michael Stark, Li-Jia Li, David Shamma, Michael Bernstein, and
Li Fei-Fei. Image retrieval using scene graphs. In CVPR, 2015.
Justin Johnson, Andrej Karpathy, and Li Fei-Fei. DenseCap: Fully convolutional localization
networks for dense captioning. In CVPR, 2016.
Justin Johnson, Bharath Hariharan, Laurens van der Maaten, Li Fei-Fei, C Lawrence Zitnick, and
Ross Girshick. CLEVR: A diagnostic dataset for compositional language and elementary visual
reasoning. In CVPR, 2017a.
Justin Johnson, Bharath Hariharan, Laurens van der Maaten, Judy Hoffman, Li Fei-Fei, C Lawrence
Zitnick, and Ross Girshick. Inferring and executing programs for visual reasoning. In ICCV,
2017b.
Ryan Kiros, Ruslan Salakhutdinov, and Richard S Zemel. Unifying visual-semantic embeddings with
multimodal neural language models. arXiv:1411.2539, 2014.
George Konidaris, Leslie Pack Kaelbling, and Tomas Lozano-Perez. From skills to symbols: Learning
symbolic representations for abstract high-level planning. JAIR, 2018.
Tejas D Kulkarni, William F Whitney, Pushmeet Kohli, and Josh Tenenbaum. Deep convolutional
inverse graphics network. In NeurIPS, 2015.
Beth Levin. English verb classes and alternations, volume 1. University of Chicago Press, 1993.
Tsung-Yi Lin, Michael Maire, Serge Belongie, James Hays, Pietro Perona, Deva Ramanan, Piotr
Dollár, and C Lawrence Zitnick. Microsoft coco: Common objects in context. In ECCV, 2014.
M Malinowski and M Fritz. A multi-world approach to question answering about real-world scenes
based on uncertain input. In NeurIPS, 2014.
Junhua Mao, Jiajing Xu, Kevin Jing, and Alan L Yuille. Training and evaluating multimodal word
embeddings with large-scale web annotated images. In NeurIPS, 2016.
David Mascharka, Philip Tran, Ryan Soklaski, and Arjun Majumdar. Transparency by design:
Closing the gap between performance and interpretability in visual reasoning. In CVPR, 2018.
Junhyuk Oh, Satinder Singh, Honglak Lee, and Pushmeet Kohli. Zero-shot task generalization with
multi-task deep reinforcement learning. In ICML, 2017.
Jeffrey Pennington, Richard Socher, and Christopher Manning. Glove: Global vectors for word
representation. In EMNLP, 2014.
Ethan Perez, Florian Strub, Harm De Vries, Vincent Dumoulin, and Aaron Courville. Film: Visual
reasoning with a general conditioning layer. In AAAI, 2018.

12
Published as a conference paper at ICLR 2019

Adam Santoro, David Raposo, David GT Barrett, Mateusz Malinowski, Razvan Pascanu, Peter
Battaglia, and Timothy Lillicrap. A simple neural network module for relational reasoning. In
NeurIPS, 2017.
Sebastian Schuster, Ranjay Krishna, Angel Chang, Li Fei-Fei, and Christopher D Manning. Generat-
ing semantically precise scene graphs from textual descriptions for improved image retrieval. In
EMNLP Workshop, 2015.
Haoyue Shi, Jiayuan Mao, Tete Xiao, Yuning Jiang, and Jian Sun. Learning visually-grounded
semantics from contrastive adversarial samples. In COLING, 2018.
N Siddharth, T. B. Paige, J.W. Meent, A. Desmaison, N. Goodman, P. Kohli, F. Wood, and P. Torr.
Learning disentangled representations with semi-supervised deep generative models. In NeurIPS,
2017.
Joseph Suarez, Justin Johnson, and Fei-Fei Li. DDRprog: A clevr differentiable dynamic reasoning
programmer. arXiv:1803.11361, 2018.
Richard S Sutton, David A McAllester, Satinder P Singh, and Yishay Mansour. Policy gradient
methods for reinforcement learning with function approximation. In NeurIPS, 2000.
Ivan Vendrov, Ryan Kiros, Sanja Fidler, and Raquel Urtasun. Order-embeddings of images and
language. In ICLR, 2016.
Ronald J Williams. Simple statistical gradient-following algorithms for connectionist reinforcement
learning. MLJ, 8(3-4):229–256, 1992.
Jiajun Wu, Joshua B Tenenbaum, and Pushmeet Kohli. Neural scene de-rendering. In CVPR, 2017.
Huijuan Xu and Kate Saenko. Ask, attend and answer: Exploring question-guided spatial attention
for visual question answering. In ECCV, 2016.
Jimei Yang, Scott E Reed, Ming-Hsuan Yang, and Honglak Lee. Weakly-supervised disentangling
with recurrent transformations for 3d view synthesis. In NeurIPS, 2015.
Zichao Yang, Xiaodong He, Jianfeng Gao, Li Deng, and Alex Smola. Stacked attention networks for
image question answering. In CVPR, 2016.
Kexin Yi, Jiajun Wu, Chuang Gan, Antonio Torralba, Pushmeet Kohli, and Joshua B Tenenbaum.
Neural-Symbolic VQA: Disentangling reasoning from vision and language understanding. In
NeurIPS, 2018.
Yukun Zhu, Ryan Kiros, Rich Zemel, Ruslan Salakhutdinov, Raquel Urtasun, Antonio Torralba, and
Sanja Fidler. Aligning books and movies: Towards story-like visual explanations by watching
movies and reading books. In CVPR, 2015.

13
Published as a conference paper at ICLR 2019

A CLEVR DOMAIN - SPECIFIC LANGUAGE AND IMPLEMENTATIONS

We first introduce the domain-specific language (DSL) designed for the CLEVR VQA dataset (John-
son et al., 2017a). Table 6 shows the available operations in the DSL, while Table 7 explains the type
system.

Operation Signature Semantics

Scene () −→ ObjectSet Return all objects in the scene.
Filter (ObjectSet, ObjConcept) −→ ObjectSet Filter out a set of objects having the
object-level concept (e.g., red) from
the input object set.
Relate (Object, RelConcept) −→ ObjectSet Filter out a set of objects that have
the relational concept (e.g., left)
with the input object.
AERelate (Object, Attribute) −→ ObjectSet (Attribute-Equality Relate) Filter
out a set of objects that have the
same attribute value (e.g., same
color) as the input object.
Intersection (ObjectSet, ObjectSet) −→ ObjectSet Return the intersection of two ob-
ject sets.
Union (ObjectSet, ObjectSet) −→ ObjectSet Return the union of two object sets.
Query (Object, Attribute) −→ ObjConcept Query the attribute (e.g., color) of
the input object.
AEQuery (Object, Object, Attribute) −→ Bool (Attribute-Equality Query) Query
if two input objects have the same
attribute value (e.g., same color).
Exist (ObjectSet) −→ Bool Query if the set is empty.
Count (ObjectSet) −→ Integer Query the number of objects in the
input set.
CLessThan (ObjectSet, ObjectSet) −→ Bool (Counting LessThan) Query if the
number of objects in the first input
set is less than the one of the second
set.
CGreaterThan (ObjectSet, ObjectSet) −→ Bool (Counting GreaterThan) Query if
the number of objects in the first
input set is greater than the one of
the second set.
CEqual (ObjectSet, ObjectSet) −→ Bool (Counting Equal) Query if the num-
ber of objects in the first input set
is the same as the one of the second
set.

Table 6: All operations in the domain-specific language for CLEVR VQA.

We note that some function takes Object as its input instead of ObjectSet. These functions require the
uniqueness of the referral object. For example, to answer the question “What’s the color of the red
object?”, there should be one and only one red object in the scene. During the program execution, the
input object set will be implicitly cast to the single object (if the set is non-empty and there is only
one object in the set). Such casting is named Unique in related works (Johnson et al., 2017b).

14
Published as a conference paper at ICLR 2019

Type Example Semantics

ObjConcept Red, Cube, etc. Object-level concepts.
Attribute Color, Shape, etc. Object-level attributes.
RelConcept Left, Front, etc. Relational concepts.
Object A single object in the scene.
ObjectSet { , } A set of objects in the scene.
Integer 0, 1, 2, · · · A single integer.
Bool True, False A single boolean value.

Table 7: The type system of the domain-specific language for CLEVR VQA.

B S EMANTIC PARSING

As shown in Appendix A, a program can be viewed as a hierarchy of operations which take concepts
as their parameters. Thus, NS-CL generates the hierarchies of latent programs in a sequence to tree
manner (Dong & Lapata, 2016). The semantic parser adopts an encoder-decoder architecture, which
contains four neural modules: (1) a bidirectional GRU encoder IEncoder (Cho et al., 2014) to
encode an input question into a fixed-length embedding, (2) an operation decoder OpDecoder that
determines the operation tokens, such as Filter, in the program based on the sentence embedding,
(3) a concept decoder ConceptDecoder that selects concepts appeared in the input question as
the parameters for certain operations (e.g., Filter takes an object-level concept parameter while
Query takes an attribute), and (4) a set of output encoders {OEncoderi } which encode the decoded
operations by OpDecoder and output the latent embedding for decoding the next operation. The
operation decoder, the concept decoder, and the output encoders work jointly and recursively to
generate the hierarchical program layout. Algorithm 1 illustrates the algorithmic outline of the
semantic parser.

Algorithm 1: The String-to-Tree Semantic Parser.

Function parse(f , {ci }):
program ← EmptyProgram();
[Link] ← OpDecoder(f );
if [Link] requires a concept parameter then
[Link] ← ConceptDecoder(f, {ci });
for i = 0, 1, · · · number of non-concept inputs of [Link] do
[Link][i] ← parse ( OEncoderi (f , [Link]) ,{ci } );
return program

The function parse takes two inputs: the current decoding state f and all concepts appeared in the
question, as a set {ci }. The parsing procedure begins with encoding the input question by IEncoder
as f0 , extracting the concept set {ci } from the input question, and invoking parse(f0 , {ci }).
The concept set {ci } is extracted using hand-coded rules. We assume that each concept (including
object-level concepts, relational concepts, and attributes) is associated with a single word in the
question. For example, the word “red” is associated with the object-level concept Red, while the
word “shape” is associated with the attribute Shape. Informally, we call these words concept words.
For a given question Q, the corresponding concept set {ci } is composed of all occurrences of the
concept words in Q. The set of concept words is known for the CLEVR dataset. For natural language
questions, one could run POS tagging to find all concept words (Andreas et al., 2016; Schuster et al.,
2015). We leave the automatic discovery of concept words as a future work (Gauthier et al., 2018).
We use the word embedding of the concept words as the representation for the concepts {ci }. Note
that, these “concept embeddings” are only for the program parsing. The visual module has separate
concept embeddings for aligning object features with concepts in the visual-semantic space.

15
Published as a conference paper at ICLR 2019

We now delve into the main function parse(f, {ci }): we first decode the root operation op of the
hierarchy by OpDecoder(f ). If op requires a concept parameter (an object-level concept, a relational
concept, or an attribute), ConceptDecoder will be invoked to choose a concept from all concepts
{ci }. Assuming op takes two non-concept inputs (e.g., the operation Intersection takes two
object sets as its input), there will be two branches for this root node. Thus, two output encoders
OEncoder0 and OEncoder1 will be applied to transform the current state f into two sub-states
f1 and f2 . parse will be recursively invoked based on f1 and f2 to generate the two branches
respectively. In the DSL, the number of non-concept inputs for any operation is at most 2.
In our implementation, the input encoder IEncoder first maps each word in the question into an
embedding space. The word embeddings are composed of two parts: a randomly initialized word
embedding of dimension 256 and a positional embedding of dimension 128 (Gehring et al., 2017).
For a concept word, its word embedding only depends on which type it belongs to (i.e. object-level,
relational or attribute). Thus, after being trained on a fixed dataset, the semantic parser can parse
questions with novel (unseen) concept words. The sequence of word embeddings is then encoded by a
two-layer GRU with a hidden dimension of 256 ∗ 2 (bidirectional). The function parse starts from the
last hidden state of the GRU, and works recursively to generate the hierarchical program layout. Both
OpDecoder and ConceptDecoder are feed-forward networks. ConceptDecoder performs
attentions over the representations of all concepts {ci } to select the concepts. Output encoders
OEncoder0 and OEncoder1 are implemented as GRU cells.
Another pre-processing of the sentence is to group consecutive object-level concept words into a
group and treat them together as a single concept, inspired by the notion of “noun phrases” in natural
languages. The computational intuition behind this grouping is that, the latent programs of CLEVR
questions usually contain multiple consecutive Filter tokens. During the program parsing and
execution, we aim to fuse all such Filters into a single Filter operation that takes multiple
concepts as its parameter.

A Running Example As a running example, consider again the question “What is the color of
the cube right of the red matte object?”. We first process the sentence (by rules) as: “What is the
<Attribute 1 (color)> of the <(ObjConcept 1 (cube)> <RelConcept 1 (right)> of the <ObjConcept
2 (red matte object)>?”. The expected parsing result of this sentence is:
Query(<Attribute 1>,
Filter(<ObjConcept 1>,
Relate(<RelConcept 1>,
Filter(<ObjConcept 2>, Scene)
)
)
).
The semantic parser encode the word embeddings with IEncoder. The last hidden state of
the GRU will be used as f0 . The word embeddings of the concept words form the set {ci } =
{Attribute 1, ObjConcept 1, RelConcept 1, ObjConcept 2}. The function parse is then invoked re-
cursively to generate the hierarchical program layout. Table 8 illustrates the decoding process
step-by-step.

C P ROGRAM E XECUTION
In this section, we present the implementation of all operations listed in Table 6. We start from the
implementation of Object-typed and ObjectSet-typed variables. Next, we discuss how to classify
objects by object-level concepts or relational concept, followed by the implementation details of all
operations.

Object-typed and ObjectSet-typed variables. We consider a scene with n objects. An Object-

typed
P variable can be represented as a vector Object of length n, where Objecti ∈ [0, 1] and
i Objecti = 1. Objecti can be interpreted as the probability that the i-th object of the scene is
being referred to. Similarly, an ObjectSet-typed variable can be represented as a vector ObjectSet
of length n, where ObjectSeti ∈ [0, 1]. ObjectSeti can be interpreted as the probability that the
i-the object is in the set. To cast an ObjectSet-typed variable ObjectSet as an Object-typed variable

16
Published as a conference paper at ICLR 2019

Step Inputs Outputs Recursive Invocation

1 f0 OpDecoder(f0 ) → Query; parse(f1 )
ConceptDecoder(f0 ) →< Attribute 1 >;
OEncoder0 (f0 , Query) → f1
2 f1 OpDecoder(f1 ) → Filter; parse(f2 )
ConceptDecoder(f1 ) →< ObjConcept 1 >;
OEncoder0 (f1 , Filter) → f2
3 f2 OpDecoder(f2 ) → Relate; parse(f3 )
ConceptDecoder(f2 ) →< RelConcept 1 >;
OEncoder0 (f2 , Relate) → f3
4 f3 OpDecoder(f3 ) → Filter; parse(f4 )
ConceptDecoder(f3 ) →< ObjConcept 2 >;
OEncoder0 (f3 , Filter) → f4
5 f4 OpDecoder(f3 ) → Scene; (End of branch.)

Table 8: A step-by-step running example of the recursive parsing procedure. The parameter {ci } is
omitted for better visualization.

Object (i.e., the Unique operation), we compute: Object = softmax(σ −1 (ObjectSet)), where
σ −1 (x) = log(x/(1 − x)) is the logit function.

Concept quantization. Denote oi as the visual representation of the i-th object, OC the set of all
object-level concepts, and A the set of all object-level attributes. Each object-level concept oc (e.g.,
Red) is associated with a vector embedding v oc and a L1-normalized vector boc of length |A|. boc
represents which attribute does this object-level concept belong to (e.g., the concept Red belongs to
the attribute Color). All attributes a ∈ A are implemented as neural operators, denoted as ua (e.g.,
uColor ). To classify the objects as being Red or not, we compute:
X a
!
hu (o i ), v Red i − γ
Pr[object i is Red] = σ bRed
a · ,
τ
a∈A

where σ denotes the Sigmoid function, h·, ·i the cosine distance between two vectors. γ and τ
are scalar constants for scaling and shifting the values of similarities. By applying this classifier
on all objects we will obtain a vector of length n, denoted as ObjClassify(Red). Similarly, such
classification can be done for relational concepts such as Left. This will result in an n × n matrix
RelClassify(Left), where RelClassify(Left)j,i is the probability that the object i is left of the
object j.
To classify whether two objects have the same attribute (e.g., have the same Color), we compute:
 Color
(oi ), uColor (oj )i − γ

hu
Pr[object i has the same Color as object j] = σ ,
τ
We can obtain a matrix AEClassify(Color) by applying this classifier on all pairs of objects, where
AEClassifier(Color)j,i is the probability that the object i and j have the same Color.

Quasi-symbolic program execution. Finally, Table 9 summarizes the implementation of all oper-
ators. In practice, all probabilities are stored in the log space for better numeric stability.

D O PTIMIZATION OF THE S EMANTIC PARSER

To tackle the optimization in a non-smooth program space, we apply an off-policy program search
process (Sutton et al., 2000) to facilitate the learning of the semantic parser. Denote P(s) as the set of
all valid programs in the CLEVR DSL for the input question s. We want to compute the gradient
w.r.t. Θs , the parameters of the semantic parser:
∇Θs = ∇Θs EP [r · log Pr[P ]],

17
Published as a conference paper at ICLR 2019

Signature Implementation
Scene() → out: ObjectSet outi := 1
Filter(in: ObjectSet, oc: ObjConcept) → outi := min(ini , ObjClassify(oc)i )
out: ObjectSet
P
Relate(in: Object, rc: RelConcept) → outi := j (inj · RelClassify(rc)j,i ))
out: ObjectSet
P
AERelate(in: Object, a: Attribute) → outi := j (inj · AEClassify(a)j,i ))
out: ObjectSet
(1) (2)
Intersection(in(1) : ObjectSet, outi := min(ini , ini )
in(2) : ObjectSet) → out: ObjectSet
(1) (2)
Union(in(1) : ObjectSet, in(2) : ObjectSet) → outi := max(ini , ini )
out: ObjectSet
Query(in: Object, a: Attribute) → P ObjClassify(oc)i · boc
a
Pr[out = oc] := i ini · P oc0
out: ObjConcept 0
oc0 ObjClassify(oc )i · ba
(1) (2)
AEQuery(in(1) : Object, in(2) : Object,
P P
b := i j (ini · inj · AEClassify(a)j,i ))
a: Attribute) → b: Bool
Exist(in: ObjectSet) → b: Bool b := maxi ini
P
Count(in: ObjectSet) → i: Integer i := i ini
P (2) P (1)
CLessThan(in(1) : ObjectSet,


b := σ ( i ini − i ini − 1 + γc )/τc
in(2) : ObjectSet) → b: Bool
(1) (2)
CGreaterThan(in(1) : ObjectSet,
P P

b := σ ( i ini − i ini − 1 + γc )/τc
in(2) : ObjectSet) → b: Bool
(1) (2)
CEqual(in(1) : ObjectSet,
P P

b := σ (−| i ini − i ini | + γc )/(γc · τc )
in(2) : ObjectSet) → b: Bool

Table 9: All operations in the domain-specific language for CLEVR VQA. γc = 0.5 and τc = 0.25
are constants for scaling and shift the probability. During inference, one can quantify all operations
as Yi et al. (2018).

where P ∼ SemanticParse(s; Θs ). In REINFORCE, we approximate this gradient via Monte Carlo

sampling.
An alternative solution is to exactly compute the gradient. Note that in the definition of the reward
r, only the set of programs Q(s) leading to the correct answer will contribute to the gradient term.
With the perception module fixed, the set Q can be efficiently determined by an off-policy exhaustive
search of all possible programs P(s). In the third stage of the curriculum learning, we search for the
set Q offline based on the quantified results of concept classification and compute the exact gradient
∇Θs . An intuitive explanation of the off-policy search is that, we enumerate all possible programs,
execute them on the visual representation, and find the ones leading to the correct answer. We use
Q(s) as the “groundtruth” program annotation for the question, to supervise the learning, instead of
running the Monte Carlo sampling-based REINFORCE.

Spurious
P program suppression. However, directly using Q(s) as the supervision by computing
` = p∈Q(S) − log Pr(p) can be problematic, due to the spuriousness or the ambiguity of the
programs. This comes from two aspects:
1) intrinsic ambiguity: two programs are different but equivalent. For example

P1: AEQuery(Color, Filter(Cube), Filter(Sphere)) and

P2: Exist(Filter(Sphere, AERelate(Color, Filter(Cube))))

are equivalent.
2) extrinsic spuriousness: one of the program is incorrect, but also leads to the correct answer in a

18
Published as a conference paper at ICLR 2019

specific scene. For example,

P1: Filter(Red, Relate(Left, Filter(Sphere))) and
P2: Filter(Red, Relate(Left, Filter(Cube)))
may refer to the same red object in a specific scene. Motivated by the REINFORCE process, to
suppress such spurious programs, we use the loss function:
X
`= stop gradient(Pr[p]) · (− log Pr[p]).
p∈Q

The corresponding gradient ∇Θs is,

 
X X
∇Θs = Pr[p] · ∇Θs (r · log Pr[P ]) = ∇Θs  r · Pr[p] .
p∈Q p∈Q

The key observation is that, given a sufficiently large set of scenes, a program can be identified
as spurious if there exists at least one scene where the program leads to a wrong answer. As the
training goes, spurious programs will get less update due to the sampling importance term Pr[p]
which weights the likelihood maximization term.

E C URRICULUM L EARNING S ETUP

During the whole training process, we gradually add more visual concepts and more complex question
examples into the model. Summarized in Figure 4(A), in general, the whole training process is split
into 3 stages. First, we only use questions from lesson 1 to let the model learn object-level visual
concepts. Second, we train the model to parse simple questions and to learn relational concepts. In
this step, we freeze the neural operators and concept embeddings of object-level concepts. Third, the
model gets trained on the full question set (lesson 3), learning to understand questions of different
complexities and various format. For the first several iterations in this step, we freeze the parameters
in the perception modules. In addition, during the training of all stages, we gradually increase the
number of objects in the scene: from 3 to 10.
We select questions for each lesson in the curriculum learning by their depth of the latent program
layout. For eaxmple, the program “Query(Shape, Filter(Red, Scene))” has the depth of 3,
while the program “Query(Shape, Filter(Cube, Relate(Left, Filter(Red, Scene))))”
has the depth of 5. Since we have fused consecutive Filter operations into a single one, the
maximum depth of all programs is 9 on the CLEVR dataset. We now present the detailed split of our
curriculum learning lessons:
For lesson 1, we use only programs of depth 3. It contains three types of questions: querying an
attribute of the object, querying the existence of a certain type of objects, count a certain type of
objects, and querying if two objects have the same attribute (e.g., of the same color). These questions
are almost about fundamental object-based visual concepts. For each image, we generate 5 questions
of lesson 1.
For lesson 2, we use programs of depth less than 5, containing a number of questions regarding
relations, such as querying the attribute of an object that is left of another object. We found that
in the original CLEVR dataset, all Relate operations are followed by a Filter operation. This
setup degenerates the performance of the learning of relational concepts such as Left. Thus, we
add a new question template into the original template set: Count(Relate( · , Filter( · ,
Scene))) (e.g., “What’s the number of objects that are left of the cube?”). For each image, we
generate 5 questions of lesson 2.
For lesson 3, we use the full CLEVR question set.
Curriculum learning is crucial for the learning of our neuro-symbolic concept learner. We found that
by removing the curriculum setup w.r.t. the number of object in the scenes, the visual perception
module will get stuck at an accuracy that is similar to a random-guess model, even if we only use
stage-1 questions. If we remove the curriculum setup w.r.t. the complexity of the programs, the joint
training of the visual perception module and the semantic parser can not converge.

19
Published as a conference paper at ICLR 2019

F A BLATION STUDY

We conduct ablation studies on the accuracy of semantic parsing, the impacts of the ImageNet pre-
training of visual perception modules, the data efficiency of our model, and the usage of object-based
representations.

F.1 S EMANTIC PARSING ACCURACY.

We evaluate how well our model recovers the underlying programs of questions. Due to the intrinsic
equivalence of different programs, we evaluate the accuracy of programs by executing them on the
ground-truth annotations of objects. Invalid or ambiguous programs are also considered as incorrect.
Our semantic parser archives > 99.9% QA accuracy on the validation split.

F.2 I MPACTS OF THE I MAGE N ET PRE - TRAINING .

The only extra supervision of the visual perception module comes from the pre-training of the
perception modules on ImageNet (Deng et al., 2009). To quantify the influence of this pre-training,
we conduct ablation experiments where we randomly initialize the perception module following He
et al. (2015). The classification accuracies of the learned concepts almost remain the same except for
Shape. The classification accuracy of Shape drops from 98.7 to 97.5 on the validation set while the
overall QA accuracy on the CLEVR dataset drops to 98.2 from 98.9. We speculate that large-scale
image recognition dataset can provide prior knowledge of shape.

F.3 DATA E FFICIENCY AND O BJECT- BASED R EPRESENTATIONS

In this section, we study whether and how the number of training samples and feature representations
affect the overall performance of various models on the CLEVR dataset. Specifically, we compare
the proposed NS-CL against two strong baselines: TbD (Mascharka et al., 2018) and MAC (Hudson
& Manning, 2018).

Baselines. For comparison, we implement two variants of the baseline models: TbD-Object and
MAC-Object. Inspired by Anderson et al. (2018), instead of using a 2D convolutional feature map,
TbD-Object and MAC-Object take a stack of object features as inputs, whose shape is k × dobj . k
is the number of objects in the scene, and dobj is the feature dimension for a single object. In our
experiments, we fix k = 12 as a constant value. If there are fewer than 12 objects in the scene, we
add “null” objects whose features are all-zero vectors.
We extract object features in the same way as NS-CL. Features are extracted from a pre-trained
ResNet-34 network before the last residual block for a feature map with high resolution. For each
object, its feature is composed of two parts: region-based (by RoI Align) and image-based features.
We concatenate them to represent each object. As discussed, the inclusion of the representation of the
full scene is essential for the inference of relative attributes such as size or spatial position on the
CLEVR domain.
TbD and MAC networks are originally designed to use image-level attention for reasoning. Thus, we
implement two more baselines: TbD-Mask and MAC-Mask. Specifically, we replace the original
attention module on images with a mask-guided attention. Denotes the union of all object masks as
M . Before the model applies the attention on the input image, we multiply the original attention map
computed by the model with this mask M . The multiplication silences the attention on pixels that are
not part of any objects.

Results. Table 3 summarizes the results. We found that TbD-Object and MAC-Object approach
show inferior results compared with the original model. We attribute this to the design of the network
architectures. Take the Relate operation (e.g., finds all objects left of a specific object x) as an
example. TbD uses a stack of dilated convolutional layers to propagate the attention from object x to
others. In TbD-Object, we replace the stack of 2D convolutions by several 1D convolution layers,
operating over the k × dobj object features. This ignores the equivalence of objects (the order of
objects should not affect the results). In contrast, MAC networks always use the attention mechanism

20
Published as a conference paper at ICLR 2019

to extract information from the image representation. This operation is invariant to the order of
objects, but is not suitable for handling quantities (e.g., counting objects).
As for TbD-Mask and MAC-Mask, although the mask-guided attention does not improve the overall
performance, we have observed noticeably faster convergence during model training. TbD-Mask and
MAC-Mask leverage the prior knowledge of object masks to facilitate the attention. Such prior has
also been verified to be effective in the original TbD model: TbD employs an attention regularization
during training, which encourages the model to attend to smaller regions.
In general, NS-CL is more data-efficient than MAC networks and TbD. Recall that NS-CL answers
questions by executing symbolic programs on the learned visual concepts. Only visual concepts (such
as Red and Left) and the interpretation of questions (how to translate questions into executable
programs) need to be learned from data. In contrast, both TbD and MAC networks need to additionally
learn to execute (implicit or explicit) programs such as counting.
For the experiments on the full CLEVR training set, we split 3,500 images (5% of the training data)
as the hold-out validation set to tune the hyperparameters and select the best model. We then apply
this model to the CLEVR validation split and report the testing performance. Our model reaches an
accuracy of 99.2% using the CLEVR training set.

G E XTENDING TO OTHER S CENE AND L ANGUAGE D OMAINS

G.1 M INECRAFT DATASET

We also extend the experiments to a new reasoning testbed: Minecraft worlds (Yi et al., 2018).
The Minecraft reasoning dataset differs from CLEVR in both visual appearance and question types.
Figure 9 gives an example instance from the dataset.

Q: What direction is the closest creature facing?

A: Left.
P: Query(Direction, FilterMost(Closest,
Filter(Creature)
))

Figure 9: An example image and a related question-answering pair from the Minecraft dataset.

Setup. Following Yi et al. (2018), we generate 10,000 Minecraft scenes using the officially open-
sourced tools by Wu et al. (2017). Each image contains 3 to 6 objects. The objects are chosen from
12 categories, with 4 different facing directions (front, back, left and right). They stand on a 2D plane.
Besides different 3D visual appearance and image contexts, the Minecraft reasoning dataset introduces
two new types of reasoning operations. We add them to our domain-specific language:

1. FilterMost(ObjectSet, Concept) → ObjectSet: Given a set of objects, finds the “most”

one. For example, FilterMost(Closest, set) locates the object in the input set that is
cloest to the camera (e.g., what is the direction of the closest animal?)
2. BelongTo(Object, ObjectSet) → Bool: Query if the input object belongs to a set.

Results. Table 10 summarizes the results and Figure 12 shows sample execution traces. We compare
our method against the NS-VQA baseline (Yi et al., 2018), which uses strong supervision for both
scene representation (e.g., object categories and positions) and program traces. In contrast, our
method learns both by looking at images and reading question-answering pairs. NS-CL outperforms
NS-VQA by 5% in overall accuracy. We attribute the inferior results of NS-VQA to its derendering
module. Because objects in the Minecraft world usually occlude with each other, the detected object
bounding boxes are inevitably noisy. During the training of the derendering module, each detected
bounding box is matched with one of the ground-truth bounding boxes and uses its class and pose
as supervision. Poorly localized bounding boxes lead to noisy labels and hurt the accuracy of the
derendering module. This further influences the overall performance of NS-VQA.

21
Published as a conference paper at ICLR 2019

Model Overall Count Exist Belong Query

NS-VQA 87.7 83.3 91.5 91.1 86.4

NS-CL 93.3 91.3 95.6 93.9 94.3

Table 10: Our model achieves comparable results on the Minecraft dataset with baselines trained by
full program annotations.

G.2 VQS DATASET

We conduct experiments on the VQS dataset (Gan et al., 2017). VQS is a subset of the VQA 1.0
dataset (Antol et al., 2015). It contains questions that can be visually grounded: each question is
associated with multiple image regions, annotated by humans as necessary for answering the question.

Q: Does this man have any pens on

him?
A: Yes.
P: Exist(Filter(Man,
Relate(Have, Filter(Pen))
))

Figure 10: An example image from the VQS dataset. The orange bounding boxes are object proposals.
On the right, we show the original question and answer in natural language, as well as the latent
program recovered by our parser. To answer this question, models are expected to attend to the man
and his pen in the pocket.

Setup. All models are trained on the first 63,509 images of the training set, and tested on the test
split. For hyper-parameter tuning and model selection, the rest 5,000 images from the training set
are used for validation. We use the multiple-choice setup for VQA: the models choose their most
confident answer from 18 candidate answers for each question.
To obtain the latent programs from natural languages, we use a pre-trained syntactic dependency
parser (Andreas et al., 2016; Schuster et al., 2015) for extracting programs and concepts that need to be
learned. A sample question and the program obtained by our parser is shown in Figure 10. The concept
embeddings are initialized by the bag of words (BoW) over the GloVe word embeddings (Pennington
et al., 2014).

Baselines. We compare our model against two representative baselines: MLP (Jabri et al., 2016)
and MAC (Hudson & Manning, 2018).
MLP is a standard baseline for visual-question answering, which treats the multiple-choice task as a
ranking problem. For a specific candidate answer, a multi-layer perceptron (MLP) model is used to
encode a tuple of the image, the question, and the candidate answer. The MLP outputs a score for
each tuple, and the answer to the question is the candidate with the highest score. We encode the
image with a ResNet-34 pre-trained on ImageNet and use BoW over the GloVe word embeddings for
the question and option encoding.
We slightly modify the MAC network for the VQS dataset. For each candidate answer, we concatenate
the question and the answer as the input to the model. The MAC model outputs a score from 0 to
1 and the answer to the question is the candidate with the highest score. The image features are
extracted from the same ResNet-34 model.

Results. Table 7 summarizes the results. NS-CL achieves comparable results with the MLP baseline
and the MAC network designed for visual reasoning. Our model also brings transparent reasoning
over natural images and language. Example execution traces generated by NS-CL are shown in
Figure 13. Besides, the symbolic reasoning process helps us to inspect the model and diagnose the
error sources. See the caption for details.

22
Published as a conference paper at ICLR 2019

H V ISUALIZATION OF EXECUTION TRACES AND VISUAL CONCEPTS

Another appealing benefit is that our reasoning model enjoys full interpretability. Figure 11, Figure 12,
and Figure 13 show NS-CL’s execution traces on CLEVR, Minecraft, and VQS, respectively. As a
side product, our system detects ambiguous and invalid programs and throws out exceptions. As an
example (Figure 11), the question “What’s the color of the cylinder?” can be ambiguous if there are
multiple cylinders or even invalid if there are no cylinders.
Figure 14 and Figure 15 include qualitative visualizations of the concepts learned from the CLEVR
and Minecraft datasets, including object categories, attributes, and relations. We choose samples
from the validation or test split of each dataset by generating queries of the corresponding concepts.
We set a threshold to filter the returned images and objects. For quantitative evaluations of the learned
concepts on the CLEVR dataset, please refer to Table 2 and Table 5.

23
Published as a conference paper at ICLR 2019

Example A. Example B.

Q: Do the cyan cylinder that is behind Q: There is a small blue object

the gray cylinder and the gray that is to the right of the small red
cylinder have the same material? matte object; what shape is it?

Concept Program Result Concept Program Result

Gray Cylinder Filter Small Red Filter

Matte Object

Behind Relate Right Relate

Cyan Cylinder Filter Small Blue Filter

Object

Gray Cylinder Filter Shape Query Cube (0.85)

Material AEQuery Yes (0.92)

Example C. Failure Case Example D. Ambiguous Program Case

Q: What is the color of the big box Q: What is the color of the big
left of the blue metal cylinder? metal object?

Concept Program Result Concept Program Result

Blue Metal Filter No such object found! Big Metal Filter
Cylinder Execution Object Execution
Abort Abort
Left Relate Color Query Ambiguous Referral!

Big Box Filter

Color: Blue ✓
Material: Rubber ✕
Color Query Shape: Cylinder ✓
Size: Small ✓

Figure 11: Visualization of the execution trace generated by our Neuro-Symbolic Concept Learner
on the CLEVR dataset. Example A and B are successful executions that generate correct answers.
In example C, the execution aborts at the first operator. To inspect the reason why the execution
engine fails to find the corresponding object, we can read out the visual representation of the object,
and locate the error source as the misclassification of the object material. Example D shows how
our symbolic execution engine can detect invalid or ambiguous programs during the execution by
performing sanity checks.

24
Published as a conference paper at ICLR 2019

Example A. Example B.
Q: Are there sheep near the wolf? Q: How many animals are closer
to the camera than the pig?

Concept Program Result Concept Program Result

Wolf Filter Pig Filter

Near Relate Closer Relate

Sheep Filter Animal Filter

Exist No (0.98) ✓ Count 1 ✓

Example C. Example D. Failure Case

Q: Which direction is the closest Q: How many pigs are there?
animal facing?

Concept Program Result Concept Program Result

Animal Filter Pig Filter

Closest FilterMost Count 1 ✕

Detected
Direction Face Right ✓ Missed

Figure 12: Exemplar execution trace generated by our Neuro-Symbolic Concept Learner on the
Minecraft reasoning dataset. Example A, B and C are successful execution. Example C demonstrates
the semantics of the FilterMost operation. Example D shows a failure case: the detection model fails
to detect a pig hiding behind the big tree.

25
Published as a conference paper at ICLR 2019

Example A. Example B.

Q: How many zebras are there?

Concept Program Result
Q: What is the sharp object on the table?
Concept Program Result
Zebra Filter

Table Filter

Count 3 ✓
On Relate

Shape Object Filter

What Query Knife (0.85) ✓

Example C. Example D.

Q: What kind of desert is plated?

Q: What are the kids doing?
Concept Program Result
Concept Program Result

Kids Filter
Desert, Plated Filter

What Query Playing_Frisbee (0.70) ✕

Kind Query Cake (0.68) ✓ Groundtruth: Playing_Baseball

Figure 13: Illustrative execution trace generated by our Neuro-Symbolic Concept Learner on the
VQS dataset. Execution traces A and B shown in the figure leads to the correct answer to the question.
Our model effectively learns visual concepts from data. The symbolic reasoning process brings
transparent execution trace and can easily handle quantities (e.g., object counting in Example A).
In Example C, although NS-CL answers the question correctly, it locates the wrong object during
reasoning: a dish instead of the cake. In Example D, our model misclassifies the sport as frisbee.

26
Published as a conference paper at ICLR 2019

Concept: Cylinder

Concept: Matte

Concept: Blue Sphere

Concept: Yellow Object Left of Cylinder

Figure 14: Concepts learned on the CLEVR dataset.

27
Published as a conference paper at ICLR 2019

Concept: Wolf

Concept: Animal That Faces Right

Concept: Closest Living Thing

Concept: Villager Closer Than Pig

Figure 15: Concepts learned on the Minecraft dataset.

28