Recent findings in multi-agent deep learning systems point towards the emergence of compositional languages. These claims are often made without exact analysis or testing of the language. In this work, we analyze the emergent language resulting from two different cooperative multi-agent game with more exact measures for compositionality. Our findings suggest that solutions found by deep learning models are often lacking the ability to reason on an abstract level therefore failing to generalize the learned knowledge to out of the training distribution examples. Strategies for testing compositional capacities and emergence of human-level concepts are discussed.
Cognitive Machine Learning (ENS - EHESS - PSL Research University - CNRS - INRIA)（认知机器学习（巴黎高等师范学院 - 社会科学高等学院 - 巴黎文理研究大学 - 法国国家科学研究中心 - 法国国家信息与自动化研究所））；Facebook AI Research（Facebook人工智能研究院）；ICREA（ICREA（加泰罗尼亚研究与高等教育机构））
Natural language allows us to refer to novel composite concepts by combining expressions denoting their parts according to systematic rules, a property known as \emph{compositionality}. In this paper, we study whether the language emerging in deep multi-agent simulations possesses a similar ability to refer to novel primitive combinations, and whether it accomplishes this feat by strategies akin to human-language compositionality. Equipped with new ways to measure compositionality in emergent languages inspired by disentanglement in representation learning, we establish three main results. Fir…
University of North Carolina at Chapel Hill（北卡罗来纳大学教堂山分校）
Deep learning models struggle with compositional generalization, i.e. the ability to recognize or generate novel combinations of observed elementary concepts. In hopes of enabling compositional generalization, various unsupervised learning algorithms have been proposed with inductive biases that aim to induce compositional structure in learned representations (e.g. disentangled representation and emergent language learning). In this work, we evaluate these unsupervised learning algorithms in terms of how well they enable compositional generalization. Specifically, our evaluation protocol focus…
University of Cambridge（剑桥大学）；MILA（蒙特利尔学习算法研究所）；McGill University（麦吉尔大学）；Polytechnique Montréal（蒙特利尔综合理工学院）；ServiceNow（ServiceNow公司）
In order to communicate, humans flatten a complex representation of ideas and their attributes into a single word or a sentence. We investigate the impact of representation learning in artificial agents by developing graph referential games. We empirically show that agents parametrized by graph neural networks develop a more compositional language compared to bag-of-words and sequence models, which allows them to systematically generalize to new combinations of familiar features.
University of California, Los Angeles（加利福尼亚大学洛杉矶分校）；Johns Hopkins University（约翰斯·霍普金斯大学）
Is analogical reasoning a task that must be learned to solve from scratch by applying deep learning models to massive numbers of reasoning problems? Or are analogies solved by computing similarities between structured representations of analogs? We address this question by comparing human performance on visual analogies created using images of familiar three-dimensional objects (cars and their subregions) with the performance of alternative computational models. Human reasoners achieved above-chance accuracy for all problem types, but made more errors in several conditions (e.g., when relevant…
Carnegie Mellon University（卡内基梅隆大学）；NASA Ames Research Center（美国国家航空航天局艾姆斯研究中心）；Boeing（波音公司）
As autonomy becomes prevalent in many applications, ranging from recommendation systems to fully autonomous vehicles, there is an increased need to provide safety guarantees for such systems. The problem is difficult, as these are large, complex systems which operate in uncertain environments, requiring data-driven machine-learning components. However, learning techniques such as Deep Neural Networks, widely used today, are inherently unpredictable and lack the theoretical foundations to provide strong assurance guarantees. We present a compositional approach for the scalable, formal verificat…
Position: A Theory of Deep Learning Must Include Compositional Sparsity
Center for Brains, Minds and Machines (CBMM), MIT, Cambridge, MA, USA（大脑、心智与机器中心（CBMM），麻省理工学院）；ETH Zurich, Zurich, Switzerland（苏黎世联邦理工学院）；University of Zurich, Zurich, Switzerland（苏黎世大学）；University of Milan, Milan, Italy（米兰大学）
Compositionality is a pivotal property of symbolic reasoning. However, how well recent neural models capture compositionality remains underexplored in the symbolic reasoning tasks. This study empirically addresses this question by systematically examining recently published pre-trained seq2seq models with a carefully controlled dataset of multi-hop arithmetic symbolic reasoning. We introduce a skill tree on compositionality in arithmetic symbolic reasoning that defines the hierarchical levels of complexity along with three compositionality dimensions: systematicity, productivity, and substitut…
University of California, Irvine（加州大学欧文分校）；Mila（米拉研究所）；Québec AI Insitute（魁北克人工智能研究所）
Significant advances have been made in artificial systems by using biological systems as a guide. However, there is often little interaction between computational models for emergent communication and biological models of the emergence of language. Many researchers in language origins and emergent communication take compositionality as their primary target for explaining how simple communication systems can become more like natural language. However, there is reason to think that compositionality is the wrong target on the biological side, and so too the wrong target on the machine-learning si…
New York University（纽约大学）；MILA（蒙特利尔学习算法研究所）；Facebook AI Research（Facebook人工智能研究院）；Google AI（谷歌人工智能）
Many recent works have discussed the propensity, or lack thereof, for emergent languages to exhibit properties of natural languages. A favorite in the literature is learning compositionality. We note that most of those works have focused on communicative bandwidth as being of primary importance. While important, it is not the only contributing factor. In this paper, we investigate the learning biases that affect the efficacy and compositionality of emergent languages. Our foremost contribution is to explore how capacity of a neural network impacts its ability to learn a compositional language.…
University of Zagreb Faculty of Electrical Engineering and Computing（萨格勒布大学电气工程与计算学院）
Neural networks have become an increasingly popular tool for solving many real-world problems. They are a general framework for differentiable optimization which includes many other machine learning approaches as special cases. In this thesis we build a category-theoretic formalism around a class of neural networks exemplified by CycleGAN. CycleGAN is a collection of neural networks, closed under composition, whose inductive bias is increased by enforcing composition invariants, i.e. cycle-consistencies. Inspired by Functorial Data Migration, we specify the interconnection of these networks us…
University of Sussex（萨塞克斯大学）；Human Interactivity and Language Lab（人类交互与语言实验室）；Faculty of Psychology（心理学院）；University of Warsaw（华沙大学）
Compositionality is an important explanatory target in emergent communication and language evolution. The vast majority of computational models of communication account for the emergence of only a very basic form of compositionality: trivial compositionality. A compositional protocol is trivially compositional if the meaning of a complex signal (e.g. blue circle) boils down to the intersection of meanings of its constituents (e.g. the intersection of the set of blue objects and the set of circles). A protocol is non-trivially compositional (NTC) if the meaning of a complex signal (e.g. biggest…
Studies of discrete languages emerging when neural agents communicate to solve a joint task often look for evidence of compositional structure. This stems for the expectation that such a structure would allow languages to be acquired faster by the agents and enable them to generalize better. We argue that these beneficial properties are only loosely connected to compositionality. In two experiments, we demonstrate that, depending on the task, non-compositional languages might show equal, or better, generalization performance and acquisition speed than compositional ones. Further research in th…
The drivers of compositionality in artificial languages that emerge when two (or more) agents play a non-visual referential game has been previously investigated using approaches based on the REINFORCE algorithm and the (Neural) Iterated Learning Model. Following the more recent introduction of the \textit{Straight-Through Gumbel-Softmax} (ST-GS) approach, this paper investigates to what extent the drivers of compositionality identified so far in the field apply in the ST-GS context and to what extent do they translate into (emergent) systematic generalisation abilities, when playing a visual …
Recently, a number of articles have argued that deep learning models such as GPT could also capture key aspects of language processing in the human mind and brain. However, I will argue that these models are not suitable as neural models of human language. Firstly, because they fail on fundamental boundary conditions, such as the amount of learning they require. This would in fact imply that the mechanisms of GPT and brain language processing are fundamentally different. Secondly, because they do not possess the logistics of access needed for compositional and productive human language process…
University of Tübingen（蒂宾根大学）；Max Planck Institute for Informatics（马克斯·普朗克信息学研究所）；University of Amsterdam（阿姆斯特丹大学）
Learning a common representation space between vision and language allows deep networks to relate objects in the image to the corresponding semantic meaning. We present a model that learns a shared Gaussian mixture representation imposing the compositionality of the text onto the visual domain without having explicit location supervision. By combining the spatial transformer with a representation learning approach we learn to split images into separately encoded patches to associate visual and textual representations in an interpretable manner. On variations of MNIST and CIFAR10, our model is …
Saarland University（萨尔大学）；Microsoft Research India（微软印度研究院）
Scaling large language models (LLMs) leads to an emergent capacity to learn in-context from example demonstrations. Despite progress, theoretical understanding of this phenomenon remains limited. We argue that in-context learning relies on recombination of compositional operations found in natural language data. We derive an information-theoretic bound showing how in-context learning abilities arise from generic next-token prediction when the pretraining distribution has sufficient amounts of compositional structure, under linguistically motivated assumptions. A second bound provides a theoret…
Mila, Université de Montréal（蒙特利尔大学米拉研究所）；CIFAR（加拿大高等研究院）
Although neural module networks have an architectural bias towards compositionality, they require gold standard layouts to generalize systematically in practice. When instead learning layouts and modules jointly, compositionality does not arise automatically and an explicit pressure is necessary for the emergence of layouts exhibiting the right structure. We propose to address this problem using iterated learning, a cognitive science theory of the emergence of compositional languages in nature that has primarily been applied to simple referential games in machine learning. Considering the layo…
We assume that a high-dimensional datum, like an image, is a compositional expression of a set of properties, with a complicated non-linear relationship between the datum and its properties. This paper proposes a factorial mixture prior for capturing latent properties, thereby adding structured compositionality to deep generative models. The prior treats a latent vector as belonging to Cartesian product of subspaces, each of which is quantized separately with a Gaussian mixture model. Some mixture components can be set to represent properties as observed random variables whenever labeled prope…
Indian Institute of Technology, Guwahati（印度理工学院古瓦哈提分校）
Generalization error defines the discriminability and the representation power of a deep model. In this work, we claim that feature space design using deep compositional function plays a significant role in generalization along with explicit and implicit regularizations. Our claims are being established with several image classification experiments. We show that the information loss due to convolution and max pooling can be marginalized with the compositional design, improving generalization performance. Also, we will show that learning rate decay acts as an implicit regularizer in deep model …
Several recent works have found the emergence of grounded compositional language in the communication protocols developed by mostly cooperative multi-agent systems when learned end-to-end to maximize performance on a downstream task. However, human populations learn to solve complex tasks involving communicative behaviors not only in fully cooperative settings but also in scenarios where competition acts as an additional external pressure for improvement. In this work, we investigate whether competition for performance from an external, similar agent team could act as a social influence that e…
Georgia Tech（佐治亚理工学院）；Oregon State University（俄勒冈州立大学）；Facebook AI Research（Facebook人工智能研究院）
Recent work has studied the emergence of language among deep reinforcement learning agents that must collaborate to solve a task. Of particular interest are the factors that cause language to be compositional -- i.e., express meaning by combining words which themselves have meaning. Evolutionary linguists have found that in addition to structural priors like those already studied in deep learning, the dynamics of transmitting language from generation to generation contribute significantly to the emergence of compositionality. In this paper, we introduce these cultural evolutionary dynamics int…
This paper explores a novel approach to achieving emergent compositional communication in multi-agent systems. We propose a training regime implementing template transfer, the idea of carrying over learned biases across contexts. In our method, a sender-receiver pair is first trained with disentangled loss functions and then the receiver is transferred to train a new sender with a standard loss. Unlike other methods (e.g. the obverter algorithm), our approach does not require imposing inductive biases on the architecture of the agents. We experimentally show the emergence of compositional comm…
Emergent communication in artificial agents has been studied to understand language evolution, as well as to develop artificial systems that learn to communicate with humans. We show that agents performing a cooperative navigation task in various gridworld environments learn an interpretable communication protocol that enables them to efficiently, and in many cases, optimally, solve the task. An analysis of the agents' policies reveals that emergent signals spatially cluster the state space, with signals referring to specific locations and spatial directions such as "left", "up", or "upper lef…
Artificial agents have been shown to learn to communicate when needed to complete a cooperative task. Some level of language structure (e.g., compositionality) has been found in the learned communication protocols. This observed structure is often the result of specific environmental pressures during training. By introducing new agents periodically to replace old ones, sequentially and within a population, we explore such a new pressure -- ease of teaching -- and show its impact on the structure of the resulting language.
We utilize a connection between compositional kernels and branching processes via Mehler's formula to study deep neural networks. This new probabilistic insight provides us a novel perspective on the mathematical role of activation functions in compositional neural networks. We study the unscaled and rescaled limits of the compositional kernels and explore the different phases of the limiting behavior, as the compositional depth increases. We investigate the memorization capacity of the compositional kernels and neural networks by characterizing the interplay among compositional depth, sample …
University of Colorado Boulder（科罗拉多大学博尔德分校）；MILA（蒙特利尔学习算法研究所）；Université de Montréal（蒙特利尔大学）
Standard methods in deep learning for natural language processing fail to capture the compositional structure of human language that allows for systematic generalization outside of the training distribution. However, human learners readily generalize in this way, e.g. by applying known grammatical rules to novel words. Inspired by work in neuroscience suggesting separate brain systems for syntactic and semantic processing, we implement a modification to standard approaches in neural machine translation, imposing an analogous separation. The novel model, which we call Syntactic Attention, subst…
University of Milan-Bicocca（米兰比可卡大学）；Consiglio Nazionale delle Ricerche (IBFM-CNR)（意大利国家研究委员会（分子生物成像与生理学研究所））
The current understanding of deep neural networks can only partially explain how input structure, network parameters and optimization algorithms jointly contribute to achieve the strong generalization power that is typically observed in many real-world applications. In order to improve the comprehension and interpretability of deep neural networks, we here introduce a novel theoretical framework based on the compositional structure of piecewise linear activation functions. By defining a direct acyclic graph representing the composition of activation patterns through the network layers, it is p…
Learning Sparse Compositional Functions with Norm-Constrained Neural Networks
Istituto Italiano di Tecnologia（意大利技术研究院）；Università degli Studi di Genova（热那亚大学）；MaLGa（MaLGa实验室）；DIBRIS（迪布里兹实验室）；CBMM（生物医学工程与机器人实验室）；Massachusetts Institute of Technology（麻省理工学院）
University of Oxford（牛津大学）；Google DeepMind（谷歌DeepMind）
This paper aims to explore the effect of prior disambiguation on neural network- based compositional models, with the hope that better semantic representations for text compounds can be produced. We disambiguate the input word vectors before they are fed into a compositional deep net. A series of evaluations shows the positive effect of prior disambiguation for such deep models.
University of Tübingen（图宾根大学）；University of Bath（巴斯大学）；Electronic Arts（艺电公司）；Siemens AG（西门子股份公司）；TU Munich（慕尼黑工业大学）；University of Bristol（布里斯托尔大学）；Royal Society（皇家学会）
Gaussian processes (GPs) are nonparametric priors over functions. Fitting a GP implies computing a posterior distribution of functions consistent with the observed data. Similarly, deep Gaussian processes (DGPs) should allow us to compute a posterior distribution of compositions of multiple functions giving rise to the observations. However, exact Bayesian inference is intractable for DGPs, motivating the use of various approximations. We show that the application of simplifying mean-field assumptions across the hierarchy leads to the layers of a DGP collapsing to near-deterministic transforma…
Sri Sathya Sai Institute of Higher Learning（斯里萨特亚赛高等研究院）
A formal description of the compositionality of neural networks is associated directly with the formal grammar-structure of the objects it seeks to represent. This formal grammar-structure specifies the kind of components that make up an object, and also the configurations they are allowed to be in. In other words, objects can be described as a parse-tree of its components -- a structure that can be seen as a candidate for building connection-patterns among neurons in neural networks. We present a formal grammar description of convolutional neural networks and capsule networks that shows how c…
Since first introduced, computer simulation has been an increasingly important tool in evolutionary linguistics. Recently, with the development of deep learning techniques, research in grounded language learning has also started to focus on facilitating the emergence of compositional languages without pre-defined elementary linguistic knowledge. In this work, we explore the emergence of compositional languages for numeric concepts in multi-agent communication systems. We demonstrate that compositional language for encoding numeric concepts can emerge through iterated learning in populations of…
As deep neural networks become more adept at traditional tasks, many of the most exciting new challenges concern multimodality---observations that combine diverse types, such as image and text. In this paper, we introduce a family of multimodal deep generative models derived from variational bounds on the evidence (data marginal likelihood). As part of our derivation we find that many previous multimodal variational autoencoders used objectives that do not correctly bound the joint marginal likelihood across modalities. We further generalize our objective to work with several types of deep gen…
Solving complex, temporally-extended tasks is a long-standing problem in reinforcement learning (RL). We hypothesize that one critical element of solving such problems is the notion of compositionality. With the ability to learn concepts and sub-skills that can be composed to solve longer tasks, i.e. hierarchical RL, we can acquire temporally-extended behaviors. However, acquiring effective yet general abstractions for hierarchical RL is remarkably challenging. In this paper, we propose to use language as the abstraction, as it provides unique compositional structure, enabling fast learning an…
Rule Extrapolation in Language Models: A Study of Compositional Generalization on OOD Prompts
University of Cambridge（剑桥大学）；AI Center, UCL（伦敦大学学院人工智能中心）；Max Planck Institute for Intelligent Systems（马克斯·普朗克智能系统研究所）；ELLIS Institute Tübingen（蒂宾根ELLIS研究所）；Tübingen AI Center（蒂宾根人工智能中心）
Lawrence Berkeley National Laboratory（劳伦斯伯克利国家实验室）；University of California, Berkeley（加州大学伯克利分校）；University of California, Irvine（加州大学欧文分校）
Hierarchy and compositionality are common latent properties in many natural and scientific datasets. Determining when a deep network's hidden activations represent hierarchy and compositionality is important both for understanding deep representation learning and for applying deep networks in domains where interpretability is crucial. However, current benchmark machine learning datasets either have little hierarchical or compositional structure, or the structure is not known. This gap impedes precise analysis of a network's representations and thus hinders development of new methods that can l…
Methods of deep learning have become increasingly popular in recent years, but they have not arrived in compositional data analysis. Imputation methods for compositional data are typically applied on additive, centered or isometric log-ratio representations of the data. Generally, methods for compositional data analysis can only be applied to observed positive entries in a data matrix. Therefore one tries to impute missing values or measurements that were below a detection limit. In this paper, a new method for imputing rounded zeros based on artificial neural networks is shown and compared wi…
Max Planck Institute for Psycholinguistics（马克斯·普朗克心理语言学研究所）；Tel Aviv University（特拉维夫大学）；University of Glasgow（格拉斯哥大学）
Emergent communication protocols among humans and artificial neural network agents do not yet share the same properties and show some critical mismatches in results. We describe three important phenomena with respect to the emergence and benefits of compositionality: ease-of-learning, generalization, and group size effects (i.e., larger groups create more systematic languages). The latter two are not fully replicated with neural agents, which hinders the use of neural emergent communication for language evolution research. We argue that one possible reason for these mismatches is that key cogn…
Lawrence Livermore National Laboratories（劳伦斯利弗莫尔国家实验室）
This paper studies the problem of post-hoc calibration of machine learning classifiers. We introduce the following desiderata for uncertainty calibration: (a) accuracy-preserving, (b) data-efficient, and (c) high expressive power. We show that none of the existing methods satisfy all three requirements, and demonstrate how Mix-n-Match calibration strategies (i.e., ensemble and composition) can help achieve remarkably better data-efficiency and expressive power while provably maintaining the classification accuracy of the original classifier. Mix-n-Match strategies are generic in the sense that…
We tackle a task where an agent learns to navigate in a 2D maze-like environment called XWORLD. In each session, the agent perceives a sequence of raw-pixel frames, a natural language command issued by a teacher, and a set of rewards. The agent learns the teacher's language from scratch in a grounded and compositional manner, such that after training it is able to correctly execute zero-shot commands: 1) the combination of words in the command never appeared before, and/or 2) the command contains new object concepts that are learned from another task but never learned from navigation. Our deep…
Center for Brains, Minds and Machines（脑、心智与机器中心）；Massachusetts Institute of Technology（麻省理工学院）
The main success stories of deep learning, starting with ImageNet, depend on deep convolutional networks, which on certain tasks perform significantly better than traditional shallow classifiers, such as support vector machines, and also better than deep fully connected networks; but what is so special about deep convolutional networks? Recent results in approximation theory proved an exponential advantage of deep convolutional networks with or without shared weights in approximating functions with hierarchical locality in their compositional structure. More recently, the hierarchical structur…
