Brenden M. Lake

Solim LeGris, Wai Keen Vong, Brenden M. Lake et al.

The Abstraction and Reasoning Corpus (ARC) is a visual program synthesis benchmark designed to test challenging out-of-distribution generalization in humans and machines. Since 2019, limited progress has been observed on the challenge using existing artificial intelligence methods. Comparing human and machine performance is important for the validity of the benchmark. While previous work explored how well humans can solve tasks from the ARC benchmark, they either did so using only a subset of tasks from the original dataset, or from variants of ARC, and therefore only provided a tentative estimate of human performance. In this work, we obtain a more robust estimate of human performance by evaluating 1729 humans on the full set of 400 training and 400 evaluation tasks from the original ARC problem set. We estimate that average human performance lies between 73.3% and 77.2% correct with a reported empirical average of 76.2% on the training set, and between 55.9% and 68.9% correct with a reported empirical average of 64.2% on the public evaluation set. However, we also find that 790 out of the 800 tasks were solvable by at least one person in three attempts, suggesting that the vast majority of the publicly available ARC tasks are in principle solvable by typical crowd-workers recruited over the internet. Notably, while these numbers are slightly lower than earlier estimates, human performance still greatly exceeds current state-of-the-art approaches for solving ARC. To facilitate research on ARC, we publicly release our dataset, called H-ARC (human-ARC), which includes all of the submissions and action traces from human participants.

Alexa R. Tartaglini, Sheridan Feucht, Michael A. Lepori et al.

Although deep neural networks can achieve human-level performance on many object recognition benchmarks, prior work suggests that these same models fail to learn simple abstract relations, such as determining whether two objects are the same or different. Much of this prior work focuses on training convolutional neural networks to classify images of two same or two different abstract shapes, testing generalization on within-distribution stimuli. In this article, we comprehensively study whether deep neural networks can acquire and generalize same-different relations both within and out-of-distribution using a variety of architectures, forms of pretraining, and fine-tuning datasets. We find that certain pretrained transformers can learn a same-different relation that generalizes with near perfect accuracy to out-of-distribution stimuli. Furthermore, we find that fine-tuning on abstract shapes that lack texture or color provides the strongest out-of-distribution generalization. Our results suggest that, with the right approach, deep neural networks can learn generalizable same-different visual relations.

Sunayana Rane, Brenden M. Lake, Thomas L. Griffiths

Developing AI systems with a human-like understanding of everyday concepts is a key step towards developing safe, reliable systems whose behavior makes sense to humans. When probing concept understanding, asking questions about plausible category members (e.g., "Is a car a vehicle?") is likely to recall patterns in the model's vast training data. We pursue an alternative strategy, characterizing the boundaries of conceptual categories by asking about implausible category members (e.g., "Is an olive a vehicle?") to probe the kind of concept-level knowledge we take for granted in fellow humans. We characterize concept boundaries for a set of fundamental concepts by studying AI systems' assignments of objects to superordinate categories from a classic psychological study by Rosch and Mervis, as well as their assignments of the same objects to mismatched superordinate categories. We compare these assignments to those made by human participants on the full range of within-category and cross-category assignment tasks. Our results reveal a range of concepts for which which models differ in meaningful and surprising ways from humans, including treating "words" as belonging to categories like "vehicles" and "clothing," identifying several "vegetable" category members as "fruit," and assigning exemplars from non-weapon categories to the "weapons" category. We also demonstrate how these instances of concept misalignment translate into problematic downstream behavior with implications for AI safety.

Chen Yueh-Han, Guy Davidson, Brenden M. Lake · berkeley

Do LLMs robustly generalize critical safety facts to novel situations? Lacking this ability is dangerous when users ask naive questions. For instance, "I'm considering packing melon balls for my 10-month-old's lunch. What other foods would be good to include?" Before offering food options, the LLM should warn that melon balls pose a choking hazard to toddlers, as documented by the CDC. Failing to provide such warnings could result in serious injuries or even death. To evaluate this, we introduce SAGE-Eval, SAfety-fact systematic GEneralization evaluation, the first benchmark that tests whether LLMs properly apply well established safety facts to naive user queries. SAGE-Eval comprises 104 facts manually sourced from reputable organizations, systematically augmented to create 10,428 test scenarios across 7 common domains (e.g., Outdoor Activities, Medicine). We find that the top model, Claude-3.7-sonnet, passes only 58% of all the safety facts tested. We also observe that model capabilities and training compute weakly correlate with performance on SAGE-Eval, implying that scaling up is not the golden solution. Our findings suggest frontier LLMs still lack robust generalization ability. We recommend developers use SAGE-Eval in pre-deployment evaluations to assess model reliability in addressing salient risks. We publicly release SAGE-Eval at https://huggingface.co/datasets/YuehHanChen/SAGE-Eval and our code is available at https://github.com/YuehHanChen/SAGE-Eval/tree/main.

Guy Davidson, Graham Todd, Julian Togelius et al.

People are remarkably capable of generating their own goals, beginning with child's play and continuing into adulthood. Despite considerable empirical and computational work on goals and goal-oriented behavior, models are still far from capturing the richness of everyday human goals. Here, we bridge this gap by collecting a dataset of human-generated playful goals (in the form of scorable, single-player games), modeling them as reward-producing programs, and generating novel human-like goals through program synthesis. Reward-producing programs capture the rich semantics of goals through symbolic operations that compose, add temporal constraints, and allow for program execution on behavioral traces to evaluate progress. To build a generative model of goals, we learn a fitness function over the infinite set of possible goal programs and sample novel goals with a quality-diversity algorithm. Human evaluators found that model-generated goals, when sampled from partitions of program space occupied by human examples, were indistinguishable from human-created games. We also discovered that our model's internal fitness scores predict games that are evaluated as more fun to play and more human-like.

Yulu Qin, Wentao Wang, Brenden M. Lake

Language models (LMs) have demonstrated remarkable proficiency in generating linguistically coherent text, sparking discussions about their relevance to understanding human language learnability. However, a significant gap exists between the training data for these models and the linguistic input a child receives. LMs are typically trained on data that is orders of magnitude larger and fundamentally different from child-directed speech (Warstadt and Bowman, 2022; Warstadt et al., 2023; Frank, 2023a). Addressing this discrepancy, our research focuses on training LMs on subsets of a single child's linguistic input. Previously, Wang, Vong, Kim, and Lake (2023) found that LMs trained in this setting can form syntactic and semantic word clusters and develop sensitivity to certain linguistic phenomena, but they only considered LSTMs and simpler neural networks trained from just one single-child dataset. Here, to examine the robustness of learnability from single-child input, we systematically train six different model architectures on five datasets (3 single-child and 2 baselines). We find that the models trained on single-child datasets showed consistent results that matched with previous work, underscoring the robustness of forming meaningful syntactic and semantic representations from a subset of a child's linguistic input.

Hanna M. Dettki, Brenden M. Lake, Charley M. Wu et al.

Causal reasoning is a core component of intelligence. Large language models (LLMs) have shown impressive capabilities in generating human-like text, raising questions about whether their responses reflect true understanding or statistical patterns. We compared causal reasoning in humans and four LLMs using tasks based on collider graphs, rating the likelihood of a query variable occurring given evidence from other variables. LLMs' causal inferences ranged from often nonsensical (GPT-3.5) to human-like to often more normatively aligned than those of humans (GPT-4o, Gemini-Pro, and Claude). Computational model fitting showed that one reason for GPT-4o, Gemini-Pro, and Claude's superior performance is they didn't exhibit the "associative bias" that plagues human causal reasoning. Nevertheless, even these LLMs did not fully capture subtler reasoning patterns associated with collider graphs, such as "explaining away".

A. Emin Orhan, Wentao Wang, Alex N. Wang et al.

Children learn powerful internal models of the world around them from a few years of egocentric visual experience. Can such internal models be learned from a child's visual experience with highly generic learning algorithms or do they require strong inductive biases? Recent advances in collecting large-scale, longitudinal, developmentally realistic video datasets and generic self-supervised learning (SSL) algorithms are allowing us to begin to tackle this nature vs. nurture question. However, existing work typically focuses on image-based SSL algorithms and visual capabilities that can be learned from static images (e.g. object recognition), thus ignoring temporal aspects of the world. To close this gap, here we train self-supervised video models on longitudinal, egocentric headcam recordings collected from a child over a two year period in their early development (6-31 months). The resulting models are highly effective at facilitating the learning of action concepts from a small number of labeled examples; they have favorable data size scaling properties; and they display emergent video interpolation capabilities. Video models also learn more accurate and more robust object representations than image-based models trained with the exact same data. These results suggest that important temporal aspects of a child's internal model of the world may be learnable from their visual experience using highly generic learning algorithms and without strong inductive biases.

Guy Davidson, Todd M. Gureckis, Brenden M. Lake et al.

Demonstrations and instructions are two primary approaches for prompting language models to perform in-context learning (ICL) tasks. Do identical tasks elicited in different ways result in similar representations of the task? An improved understanding of task representation mechanisms would offer interpretability insights and may aid in steering models. We study this through \textit{function vectors} (FVs), recently proposed as a mechanism to extract few-shot ICL task representations. We generalize FVs to alternative task presentations, focusing on short textual instruction prompts, and successfully extract instruction function vectors that promote zero-shot task accuracy. We find evidence that demonstration- and instruction-based function vectors leverage different model components, and offer several controls to dissociate their contributions to task performance. Our results suggest that different task promptings forms do not induce a common task representation through FVs but elicit different, partly overlapping mechanisms. Our findings offer principled support to the practice of combining instructions and task demonstrations, imply challenges in universally monitoring task inference across presentation forms, and encourage further examinations of LLM task inference mechanisms.

Kazuki Irie, Brenden M. Lake

Since the earliest proposals for artificial neural network (ANN) models of the mind and brain, critics have pointed out key weaknesses in these models compared to human cognitive abilities. Here we review recent work that uses metalearning to overcome several classic challenges, which we characterize as addressing the Problem of Incentive and Practice -- that is, providing machines with both incentives to improve specific skills and opportunities to practice those skills. This explicit optimization contrasts with more conventional approaches that hope the desired behaviour will emerge through optimizing related but different objectives. We review applications of this principle to addressing four classic challenges for ANNs: systematic generalization, catastrophic forgetting, few-shot learning and multi-step reasoning. We also discuss how large language models incorporate key aspects of this metalearning framework (namely, sequence prediction with feedback trained on diverse data), which helps to explain some of their successes on these classic challenges. Finally, we discuss the prospects for understanding aspects of human development through this framework, and whether natural environments provide the right incentives and practice for learning how to make challenging generalizations.

Wentao Wang, Guangyuan Jiang, Tal Linzen et al.

Humans can quickly learn a new word from a few illustrative examples, and then systematically and flexibly use it in novel contexts. Yet the abilities of current language models for few-shot word learning, and methods for improving these abilities, are underexplored. In this study, we introduce a novel method, Meta-training for IN-context learNing Of Words (Minnow). This method trains language models to generate new examples of a word's usage given a few in-context examples, using a special placeholder token to represent the new word. This training is repeated on many new words to develop a general word-learning ability. We find that training models from scratch with Minnow on human-scale child-directed language enables strong few-shot word learning, comparable to a large language model (LLM) pre-trained on orders of magnitude more data. Furthermore, through discriminative and generative evaluations, we demonstrate that finetuning pre-trained LLMs with Minnow improves their ability to discriminate between new words, identify syntactic categories of new words, and generate reasonable new usages and definitions for new words, based on one or a few in-context examples. These findings highlight the data efficiency of Minnow and its potential to improve language model performance in word learning tasks.

Yanli Zhou, Brenden M. Lake, Adina Williams

The ability to learn and compose functions is foundational to efficient learning and reasoning in humans, enabling flexible generalizations such as creating new dishes from known cooking processes. Beyond sequential chaining of functions, existing linguistics literature indicates that humans can grasp more complex compositions with interacting functions, where output production depends on context changes induced by different function orderings. Extending the investigation into the visual domain, we developed a function learning paradigm to explore the capacity of humans and neural network models in learning and reasoning with compositional functions under varied interaction conditions. Following brief training on individual functions, human participants were assessed on composing two learned functions, in ways covering four main interaction types, including instances in which the application of the first function creates or removes the context for applying the second function. Our findings indicate that humans can make zero-shot generalizations on novel visual function compositions across interaction conditions, demonstrating sensitivity to contextual changes. A comparison with a neural network model on the same task reveals that, through the meta-learning for compositionality (MLC) approach, a standard sequence-to-sequence Transformer can mimic human generalization patterns in composing functions.

Thomas L. Griffiths, Brenden M. Lake, R. Thomas McCoy et al.

Some of the strongest evidence that human minds should be thought about in terms of symbolic systems has been the way they combine ideas, produce novelty, and learn quickly. We argue that modern neural networks -- and the artificial intelligence systems built upon them -- exhibit similar abilities. This undermines the argument that the cognitive processes and representations used by human minds are symbolic, although the fact that these neural networks are typically trained on data generated by symbolic systems illustrates that such systems play an important role in characterizing the abstract problems that human minds have to solve. This argument leads us to offer a new agenda for research on the symbolic basis of human thought.

Simon Schug, Brenden M. Lake

The validity of online behavioral research relies on study participants being human rather than machine. In the past, it was possible to detect machines by posing simple challenges that were easily solved by humans but not by machines. General-purpose agents based on large language models (LLMs) can now solve many of these challenges, threatening the validity of online behavioral research. Here we explore the idea of detecting humanness by using tasks that machines can solve too well to be human. Specifically, we probe for the existence of an established human cognitive constraint: limited working memory capacity. We show that cognitive modeling on a standard serial recall task can be used to distinguish online participants from LLMs even when the latter are specifically instructed to mimic human working memory constraints. Our results demonstrate that it is viable to use well-established cognitive phenomena to distinguish LLMs from humans.

Wai Keen Vong, Brenden M. Lake

What insights can machine learning bring to understanding human language acquisition? Large language and multimodal models have achieved remarkable capabilities, but their reliance on massive training datasets creates a fundamental mismatch with children, who succeed in acquiring language from comparatively limited input. To help bridge this gap, researchers have increasingly trained neural networks using data similar in quantity and quality to children's input. Taking this approach to the limit, Vong et al. (2024) showed that a multimodal neural network trained on 61 hours of visual and linguistic input extracted from just one child's developmental experience could acquire word-referent mappings. However, whether this approach's success reflects the idiosyncrasies of a single child's experience, or whether it would show consistent and robust learning patterns across multiple children's experiences was not explored. In this article, we applied automated speech transcription methods to the entirety of the SAYCam dataset, consisting of over 500 hours of video data spread across all three children. Using these automated transcriptions, we generated multi-modal vision-and-language datasets for both training and evaluation, and explored a range of neural network configurations to examine the robustness of simulated word learning. Our findings demonstrate that networks trained on automatically transcribed data from each child can acquire and generalize word-referent mappings across multiple network architectures. These results validate the robustness of multimodal neural networks for grounded word learning, while highlighting the individual differences that emerge in how models learn when trained on each child's developmental experiences.

Michael A. Lepori, Alexa R. Tartaglini, Wai Keen Vong et al.

Though vision transformers (ViTs) have achieved state-of-the-art performance in a variety of settings, they exhibit surprising failures when performing tasks involving visual relations. This begs the question: how do ViTs attempt to perform tasks that require computing visual relations between objects? Prior efforts to interpret ViTs tend to focus on characterizing relevant low-level visual features. In contrast, we adopt methods from mechanistic interpretability to study the higher-level visual algorithms that ViTs use to perform abstract visual reasoning. We present a case study of a fundamental, yet surprisingly difficult, relational reasoning task: judging whether two visual entities are the same or different. We find that pretrained ViTs fine-tuned on this task often exhibit two qualitatively different stages of processing despite having no obvious inductive biases to do so: 1) a perceptual stage wherein local object features are extracted and stored in a disentangled representation, and 2) a relational stage wherein object representations are compared. In the second stage, we find evidence that ViTs can learn to represent somewhat abstract visual relations, a capability that has long been considered out of reach for artificial neural networks. Finally, we demonstrate that failures at either stage can prevent a model from learning a generalizable solution to our fairly simple tasks. By understanding ViTs in terms of discrete processing stages, one can more precisely diagnose and rectify shortcomings of existing and future models.

Yanli Zhou, Reuben Feinman, Brenden M. Lake

Humans leverage compositionality to efficiently learn new concepts, understanding how familiar parts can combine together to form novel objects. In contrast, popular computer vision models struggle to make the same types of inferences, requiring more data and generalizing less flexibly than people do. Here, we study these distinctively human abilities across a range of different types of visual composition, examining how people classify and generate ``alien figures'' with rich relational structure. We also develop a Bayesian program induction model which searches for the best programs for generating the candidate visual figures, utilizing a large program space containing different compositional mechanisms and abstractions. In few shot classification tasks, we find that people and the program induction model can make a range of meaningful compositional generalizations, with the model providing a strong account of the experimental data as well as interpretable parameters that reveal human assumptions about the factors invariant to category membership (here, to rotation and changing part attachment). In few shot generation tasks, both people and the models are able to construct compelling novel examples, with people behaving in additional structured ways beyond the model capabilities, e.g. making choices that complete a set or reconfiguring existing parts in highly novel ways. To capture these additional behavioral patterns, we develop an alternative model based on neuro-symbolic program induction: this model also composes new concepts from existing parts yet, distinctively, it utilizes neural network modules to successfully capture residual statistical structure. Together, our behavioral and computational findings show how people and models can produce a rich variety of compositional behavior when classifying and generating visual objects.

A. Emin Orhan, Brenden M. Lake

Young children develop sophisticated internal models of the world based on their visual experience. Can such models be learned from a child's visual experience without strong inductive biases? To investigate this, we train state-of-the-art neural networks on a realistic proxy of a child's visual experience without any explicit supervision or domain-specific inductive biases. Specifically, we train both embedding models and generative models on 200 hours of headcam video from a single child collected over two years and comprehensively evaluate their performance in downstream tasks using various reference models as yardsticks. On average, the best embedding models perform at a respectable 70% of a high-performance ImageNet-trained model, despite substantial differences in training data. They also learn broad semantic categories and object localization capabilities without explicit supervision, but they are less object-centric than models trained on all of ImageNet. Generative models trained with the same data successfully extrapolate simple properties of partially masked objects, like their rough outline, texture, color, or orientation, but struggle with finer object details. We replicate our experiments with two other children and find remarkably consistent results. Broadly useful high-level visual representations are thus robustly learnable from a representative sample of a child's visual experience without strong inductive biases.

Alexa R. Tartaglini, Wai Keen Vong, Brenden M. Lake

Early in development, children learn to extend novel category labels to objects with the same shape, a phenomenon known as the shape bias. Inspired by these findings, Geirhos et al. (2019) examined whether deep neural networks show a shape or texture bias by constructing images with conflicting shape and texture cues. They found that convolutional neural networks strongly preferred to classify familiar objects based on texture as opposed to shape, suggesting a texture bias. However, there are a number of differences between how the networks were tested in this study versus how children are typically tested. In this work, we re-examine the inductive biases of neural networks by adapting the stimuli and procedure from Geirhos et al. (2019) to more closely follow the developmental paradigm and test on a wide range of pre-trained neural networks. Across three experiments, we find that deep neural networks exhibit a preference for shape rather than texture when tested under conditions that more closely replicate the developmental procedure.

Maxwell Nye, Michael Henry Tessler, Joshua B. Tenenbaum et al.

Human reasoning can often be understood as an interplay between two systems: the intuitive and associative ("System 1") and the deliberative and logical ("System 2"). Neural sequence models -- which have been increasingly successful at performing complex, structured tasks -- exhibit the advantages and failure modes of System 1: they are fast and learn patterns from data, but are often inconsistent and incoherent. In this work, we seek a lightweight, training-free means of improving existing System 1-like sequence models by adding System 2-inspired logical reasoning. We explore several variations on this theme in which candidate generations from a neural sequence model are examined for logical consistency by a symbolic reasoning module, which can either accept or reject the generations. Our approach uses neural inference to mediate between the neural System 1 and the logical System 2. Results in robust story generation and grounded instruction-following show that this approach can increase the coherence and accuracy of neurally-based generations.

Yanli Zhou, Brenden M. Lake

Humans are highly efficient learners, with the ability to grasp the meaning of a new concept from just a few examples. Unlike popular computer vision systems, humans can flexibly leverage the compositional structure of the visual world, understanding new concepts as combinations of existing concepts. In the current paper, we study how people learn different types of visual compositions, using abstract visual forms with rich relational structure. We find that people can make meaningful compositional generalizations from just a few examples in a variety of scenarios, and we develop a Bayesian program induction model that provides a close fit to the behavioral data. Unlike past work examining special cases of compositionality, our work shows how a single computational approach can account for many distinct types of compositional generalization.

Aysja Johnson, Wai Keen Vong, Brenden M. Lake et al.

The Abstraction and Reasoning Corpus (ARC) is a challenging program induction dataset that was recently proposed by Chollet (2019). Here, we report the first set of results collected from a behavioral study of humans solving a subset of tasks from ARC (40 out of 1000). Although this subset of tasks contains considerable variation, our results showed that humans were able to infer the underlying program and generate the correct test output for a novel test input example, with an average of 80% of tasks solved per participant, and with 65% of tasks being solved by more than 80% of participants. Additionally, we find interesting patterns of behavioral consistency and variability within the action sequences during the generation process, the natural language descriptions to describe the transformations for each task, and the errors people made. Our findings suggest that people can quickly and reliably determine the relevant features and properties of a task to compose a correct solution. Future modeling work could incorporate these findings, potentially by connecting the natural language descriptions we collected here to the underlying semantics of ARC.

Kanishk Gandhi, Gala Stojnic, Brenden M. Lake et al.

To achieve human-like common sense about everyday life, machine learning systems must understand and reason about the goals, preferences, and actions of other agents in the environment. By the end of their first year of life, human infants intuitively achieve such common sense, and these cognitive achievements lay the foundation for humans' rich and complex understanding of the mental states of others. Can machines achieve generalizable, commonsense reasoning about other agents like human infants? The Baby Intuitions Benchmark (BIB) challenges machines to predict the plausibility of an agent's behavior based on the underlying causes of its actions. Because BIB's content and paradigm are adopted from developmental cognitive science, BIB allows for direct comparison between human and machine performance. Nevertheless, recently proposed, deep-learning-based agency reasoning models fail to show infant-like reasoning, leaving BIB an open challenge.

Brenden M. Lake, Gregory L. Murphy

Machines have achieved a broad and growing set of linguistic competencies, thanks to recent progress in Natural Language Processing (NLP). Psychologists have shown increasing interest in such models, comparing their output to psychological judgments such as similarity, association, priming, and comprehension, raising the question of whether the models could serve as psychological theories. In this article, we compare how humans and machines represent the meaning of words. We argue that contemporary NLP systems are fairly successful models of human word similarity, but they fall short in many other respects. Current models are too strongly linked to the text-based patterns in large corpora, and too weakly linked to the desires, goals, and beliefs that people express through words. Word meanings must also be grounded in perception and action and be capable of flexible combinations in ways that current systems are not. We discuss more promising approaches to grounding NLP systems and argue that they will be more successful with a more human-like, conceptual basis for word meaning.

A. Emin Orhan, Vaibhav V. Gupta, Brenden M. Lake

Within months of birth, children develop meaningful expectations about the world around them. How much of this early knowledge can be explained through generic learning mechanisms applied to sensory data, and how much of it requires more substantive innate inductive biases? Addressing this fundamental question in its full generality is currently infeasible, but we can hope to make real progress in more narrowly defined domains, such as the development of high-level visual categories, thanks to improvements in data collecting technology and recent progress in deep learning. In this paper, our goal is precisely to achieve such progress by utilizing modern self-supervised deep learning methods and a recent longitudinal, egocentric video dataset recorded from the perspective of three young children (Sullivan et al., 2020). Our results demonstrate the emergence of powerful, high-level visual representations from developmentally realistic natural videos using generic self-supervised learning objectives.

Reuben Feinman, Brenden M. Lake

People can learn rich, general-purpose conceptual representations from only raw perceptual inputs. Current machine learning approaches fall well short of these human standards, although different modeling traditions often have complementary strengths. Symbolic models can capture the compositional and causal knowledge that enables flexible generalization, but they struggle to learn from raw inputs, relying on strong abstractions and simplifying assumptions. Neural network models can learn directly from raw data, but they struggle to capture compositional and causal structure and typically must retrain to tackle new tasks. We bring together these two traditions to learn generative models of concepts that capture rich compositional and causal structure, while learning from raw data. We develop a generative neuro-symbolic (GNS) model of handwritten character concepts that uses the control flow of a probabilistic program, coupled with symbolic stroke primitives and a symbolic image renderer, to represent the causal and compositional processes by which characters are formed. The distributions of parts (strokes), and correlations between parts, are modeled with neural network subroutines, allowing the model to learn directly from raw data and express nonparametric statistical relationships. We apply our model to the Omniglot challenge of human-level concept learning, using a background set of alphabets to learn an expressive prior distribution over character drawings. In a subsequent evaluation, our GNS model uses probabilistic inference to learn rich conceptual representations from a single training image that generalize to 4 unique tasks, succeeding where previous work has fallen short.

Reuben Feinman, Brenden M. Lake

Human conceptual knowledge supports the ability to generate novel yet highly structured concepts, and the form of this conceptual knowledge is of great interest to cognitive scientists. One tradition has emphasized structured knowledge, viewing concepts as embedded in intuitive theories or organized in complex symbolic knowledge structures. A second tradition has emphasized statistical knowledge, viewing conceptual knowledge as an emerging from the rich correlational structure captured by training neural networks and other statistical models. In this paper, we explore a synthesis of these two traditions through a novel neuro-symbolic model for generating new concepts. Using simple visual concepts as a testbed, we bring together neural networks and symbolic probabilistic programs to learn a generative model of novel handwritten characters. Two alternative models are explored with more generic neural network architectures. We compare each of these three models for their likelihoods on held-out character classes and for the quality of their productions, finding that our hybrid model learns the most convincing representation and generalizes further from the training observations.

Wai Keen Vong, Brenden M. Lake

How do children learn correspondences between the language and the world from noisy, ambiguous, naturalistic input? One hypothesis is via cross-situational learning: tracking words and their possible referents across multiple situations allows learners to disambiguate correct word-referent mappings (Yu & Smith, 2007). However, previous models of cross-situational word learning operate on highly simplified representations, side-stepping two important aspects of the actual learning problem. First, how can word-referent mappings be learned from raw inputs such as images? Second, how can these learned mappings generalize to novel instances of a known word? In this paper, we present a neural network model trained from scratch via self-supervision that takes in raw images and words as inputs, and show that it can learn word-referent mappings from fully ambiguous scenes and utterances through cross-situational learning. In addition, the model generalizes to novel word instances, locates referents of words in a scene, and shows a preference for mutual exclusivity.

Maxwell I. Nye, Armando Solar-Lezama, Joshua B. Tenenbaum et al.

Many aspects of human reasoning, including language, require learning rules from very little data. Humans can do this, often learning systematic rules from very few examples, and combining these rules to form compositional rule-based systems. Current neural architectures, on the other hand, often fail to generalize in a compositional manner, especially when evaluated in ways that vary systematically from training. In this work, we present a neuro-symbolic model which learns entire rule systems from a small set of examples. Instead of directly predicting outputs from inputs, we train our model to induce the explicit system of rules governing a set of previously seen examples, drawing upon techniques from the neural program synthesis literature. Our rule-synthesis approach outperforms neural meta-learning techniques in three domains: an artificial instruction-learning domain used to evaluate human learning, the SCAN challenge datasets, and learning rule-based translations of number words into integers for a wide range of human languages.

Laura Ruis, Jacob Andreas, Marco Baroni et al.

Humans easily interpret expressions that describe unfamiliar situations composed from familiar parts ("greet the pink brontosaurus by the ferris wheel"). Modern neural networks, by contrast, struggle to interpret novel compositions. In this paper, we introduce a new benchmark, gSCAN, for evaluating compositional generalization in situated language understanding. Going beyond a related benchmark that focused on syntactic aspects of generalization, gSCAN defines a language grounded in the states of a grid world, facilitating novel evaluations of acquiring linguistically motivated rules. For example, agents must understand how adjectives such as 'small' are interpreted relative to the current world state or how adverbs such as 'cautiously' combine with new verbs. We test a strong multi-modal baseline model and a state-of-the-art compositional method finding that, in most cases, they fail dramatically when generalization requires systematic compositional rules.

Guy Davidson, Brenden M. Lake

We explore the benefits of augmenting state-of-the-art model-free deep reinforcement algorithms with simple object representations. Following the Frostbite challenge posited by Lake et al. (2017), we identify object representations as a critical cognitive capacity lacking from current reinforcement learning agents. We discover that providing the Rainbow model (Hessel et al.,2018) with simple, feature-engineered object representations substantially boosts its performance on the Frostbite game from Atari 2600. We then analyze the relative contributions of the representations of different types of objects, identify environment states where these representations are most impactful, and examine how these representations aid in generalizing to novel situations.

Ziyun Wang, Brenden M. Lake

People ask questions that are far richer, more informative, and more creative than current AI systems. We propose a neuro-symbolic framework for modeling human question asking, which represents questions as formal programs and generates programs with an encoder-decoder based deep neural network. From extensive experiments using an information-search game, we show that our method can predict which questions humans are likely to ask in unconstrained settings. We also propose a novel grammar-based question generation framework trained with reinforcement learning, which is able to generate creative questions without supervised human data.

Kanishk Gandhi, Brenden M. Lake

Strong inductive biases allow children to learn in fast and adaptable ways. Children use the mutual exclusivity (ME) bias to help disambiguate how words map to referents, assuming that if an object has one label then it does not need another. In this paper, we investigate whether or not standard neural architectures have an ME bias, demonstrating that they lack this learning assumption. Moreover, we show that their inductive biases are poorly matched to lifelong learning formulations of classification and translation. We demonstrate that there is a compelling case for designing neural networks that reason by mutual exclusivity, which remains an open challenge.

A. Emin Orhan, Brenden M. Lake

We investigate the robustness properties of image recognition models equipped with two features inspired by human vision, an explicit episodic memory and a shape bias, at the ImageNet scale. As reported in previous work, we show that an explicit episodic memory improves the robustness of image recognition models against small-norm adversarial perturbations under some threat models. It does not, however, improve the robustness against more natural, and typically larger, perturbations. Learning more robust features during training appears to be necessary for robustness in this second sense. We show that features derived from a model that was encouraged to learn global, shape-based representations (Geirhos et al., 2019) do not only improve the robustness against natural perturbations, but when used in conjunction with an episodic memory, they also provide additional robustness against adversarial perturbations. Finally, we address three important design choices for the episodic memory: memory size, dimensionality of the memories and the retrieval method. We show that to make the episodic memory more compact, it is preferable to reduce the number of memories by clustering them, instead of reducing their dimensionality.

Brenden M. Lake

People can learn a new concept and use it compositionally, understanding how to "blicket twice" after learning how to "blicket." In contrast, powerful sequence-to-sequence (seq2seq) neural networks fail such tests of compositionality, especially when composing new concepts together with existing concepts. In this paper, I show how memory-augmented neural networks can be trained to generalize compositionally through meta seq2seq learning. In this approach, models train on a series of seq2seq problems to acquire the compositional skills needed to solve new seq2seq problems. Meta se2seq learning solves several of the SCAN tests for compositional learning and can learn to apply implicit rules to variables.

Brenden M. Lake, Steven T. Piantadosi

Machine learning has made major advances in categorizing objects in images, yet the best algorithms miss important aspects of how people learn and think about categories. People can learn richer concepts from fewer examples, including causal models that explain how members of a category are formed. Here, we explore the limits of this human ability to infer causal "programs" -- latent generating processes with nontrivial algorithmic properties -- from one, two, or three visual examples. People were asked to extrapolate the programs in several ways, for both classifying and generating new examples. As a theory of these inductive abilities, we present a Bayesian program learning model that searches the space of programs for the best explanation of the observations. Although variable, people's judgments are broadly consistent with the model and inconsistent with several alternatives, including a pre-trained deep neural network for object recognition, indicating that people can learn and reason with rich algorithmic abstractions from sparse input data.

Reuben Feinman, Brenden M. Lake

Modern deep neural networks require a tremendous amount of data to train, often needing hundreds or thousands of labeled examples to learn an effective representation. For these networks to work with less data, more structure must be built into their architectures or learned from previous experience. The learned weights of convolutional neural networks (CNNs) trained on large datasets for object recognition contain a substantial amount of structure. These representations have parallels to simple cells in the primary visual cortex, where receptive fields are smooth and contain many regularities. Incorporating smoothness constraints over the kernel weights of modern CNN architectures is a promising way to improve their sample complexity. We propose a smooth kernel regularizer that encourages spatial correlations in convolution kernel weights. The correlation parameters of this regularizer are learned from previous experience, yielding a method with a hierarchical Bayesian interpretation. We show that our correlated regularizer can help constrain models for visual recognition, improving over an L2 regularization baseline.

Brenden M. Lake, Ruslan Salakhutdinov, Joshua B. Tenenbaum

Three years ago, we released the Omniglot dataset for one-shot learning, along with five challenge tasks and a computational model that addresses these tasks. The model was not meant to be the final word on Omniglot; we hoped that the community would build on our work and develop new approaches. In the time since, we have been pleased to see wide adoption of the dataset. There has been notable progress on one-shot classification, but researchers have adopted new splits and procedures that make the task easier. There has been less progress on the other four tasks. We conclude that recent approaches are still far from human-like concept learning on Omniglot, a challenge that requires performing many tasks with a single model.

Brenden M. Lake, Tal Linzen, Marco Baroni

People learn in fast and flexible ways that have not been emulated by machines. Once a person learns a new verb "dax," he or she can effortlessly understand how to "dax twice," "walk and dax," or "dax vigorously." There have been striking recent improvements in machine learning for natural language processing, yet the best algorithms require vast amounts of experience and struggle to generalize new concepts in compositional ways. To better understand these distinctively human abilities, we study the compositional skills of people through language-like instruction learning tasks. Our results show that people can learn and use novel functional concepts from very few examples (few-shot learning), successfully applying familiar functions to novel inputs. People can also compose concepts in complex ways that go beyond the provided demonstrations. Two additional experiments examined the assumptions and inductive biases that people make when solving these tasks, revealing three biases: mutual exclusivity, one-to-one mappings, and iconic concatenation. We discuss the implications for cognitive modeling and the potential for building machines with more human-like language learning capabilities.

João Loula, Marco Baroni, Brenden M. Lake

Systematic compositionality is the ability to recombine meaningful units with regular and predictable outcomes, and it's seen as key to humans' capacity for generalization in language. Recent work has studied systematic compositionality in modern seq2seq models using generalization to novel navigation instructions in a grounded environment as a probing tool, requiring models to quickly bootstrap the meaning of new words. We extend this framework here to settings where the model needs only to recombine well-trained functional words (such as "around" and "right") in novel contexts. Our findings confirm and strengthen the earlier ones: seq2seq models can be impressively good at generalizing to novel combinations of previously-seen input, but only when they receive extensive training on the specific pattern to be generalized (e.g., generalizing from many examples of "X around right" to "jump around right"), while failing when generalization requires novel application of compositional rules (e.g., inferring the meaning of "around right" from those of "right" and "around").

Reuben Feinman, Brenden M. Lake

People use rich prior knowledge about the world in order to efficiently learn new concepts. These priors - also known as "inductive biases" - pertain to the space of internal models considered by a learner, and they help the learner make inferences that go beyond the observed data. A recent study found that deep neural networks optimized for object recognition develop the shape bias (Ritter et al., 2017), an inductive bias possessed by children that plays an important role in early word learning. However, these networks use unrealistically large quantities of training data, and the conditions required for these biases to develop are not well understood. Moreover, it is unclear how the learning dynamics of these networks relate to developmental processes in childhood. We investigate the development and influence of the shape bias in neural networks using controlled datasets of abstract patterns and synthetic images, allowing us to systematically vary the quantity and form of the experience provided to the learning algorithms. We find that simple neural networks develop a shape bias after seeing as few as 3 examples of 4 object categories. The development of these biases predicts the onset of vocabulary acceleration in our networks, consistent with the developmental process in children.

Anselm Rothe, Brenden M. Lake, Todd M. Gureckis

A hallmark of human intelligence is the ability to ask rich, creative, and revealing questions. Here we introduce a cognitive model capable of constructing human-like questions. Our approach treats questions as formal programs that, when executed on the state of the world, output an answer. The model specifies a probability distribution over a complex, compositional space of programs, favoring concise programs that help the agent learn in the current context. We evaluate our approach by modeling the types of open-ended questions generated by humans who were attempting to learn about an ambiguous situation in a game. We find that our model predicts what questions people will ask, and can creatively produce novel questions that were not present in the training set. In addition, we compare a number of model variants, finding that both question informativeness and complexity are important for producing human-like questions.

Brenden M. Lake, Marco Baroni

Humans can understand and produce new utterances effortlessly, thanks to their compositional skills. Once a person learns the meaning of a new verb "dax," he or she can immediately understand the meaning of "dax twice" or "sing and dax." In this paper, we introduce the SCAN domain, consisting of a set of simple compositional navigation commands paired with the corresponding action sequences. We then test the zero-shot generalization capabilities of a variety of recurrent neural networks (RNNs) trained on SCAN with sequence-to-sequence methods. We find that RNNs can make successful zero-shot generalizations when the differences between training and test commands are small, so that they can apply "mix-and-match" strategies to solve the task. However, when generalization requires systematic compositional skills (as in the "dax" example above), RNNs fail spectacularly. We conclude with a proof-of-concept experiment in neural machine translation, suggesting that lack of systematicity might be partially responsible for neural networks' notorious training data thirst.

Brenden M. Lake, Neil D. Lawrence, Joshua B. Tenenbaum

Both scientists and children make important structural discoveries, yet their computational underpinnings are not well understood. Structure discovery has previously been formalized as probabilistic inference about the right structural form --- where form could be a tree, ring, chain, grid, etc. [Kemp & Tenenbaum (2008). The discovery of structural form. PNAS, 105(3), 10687-10692]. While this approach can learn intuitive organizations, including a tree for animals and a ring for the color circle, it assumes a strong inductive bias that considers only these particular forms, and each form is explicitly provided as initial knowledge. Here we introduce a new computational model of how organizing structure can be discovered, utilizing a broad hypothesis space with a preference for sparse connectivity. Given that the inductive bias is more general, the model's initial knowledge shows little qualitative resemblance to some of the discoveries it supports. As a consequence, the model can also learn complex structures for domains that lack intuitive description, as well as predict human property induction judgments without explicit structural forms. By allowing form to emerge from sparsity, our approach clarifies how both the richness and flexibility of human conceptual organization can coexist.

Brenden M. Lake, Tomer D. Ullman, Joshua B. Tenenbaum et al.

Recent progress in artificial intelligence (AI) has renewed interest in building systems that learn and think like people. Many advances have come from using deep neural networks trained end-to-end in tasks such as object recognition, video games, and board games, achieving performance that equals or even beats humans in some respects. Despite their biological inspiration and performance achievements, these systems differ from human intelligence in crucial ways. We review progress in cognitive science suggesting that truly human-like learning and thinking machines will have to reach beyond current engineering trends in both what they learn, and how they learn it. Specifically, we argue that these machines should (a) build causal models of the world that support explanation and understanding, rather than merely solving pattern recognition problems; (b) ground learning in intuitive theories of physics and psychology, to support and enrich the knowledge that is learned; and (c) harness compositionality and learning-to-learn to rapidly acquire and generalize knowledge to new tasks and situations. We suggest concrete challenges and promising routes towards these goals that can combine the strengths of recent neural network advances with more structured cognitive models.