Deep reinforcement learning (RL) is emerging as a viable strategy for automated cyber defense (ACD). The traditional RL approach represents networks as a list of computers in various states of safety or threat. Unfortunately, these models are forced to overfit to specific network topologies, rendering them ineffective when faced with even small environmental perturbations. In this work, we frame ACD as a two-player context-based partially observable Markov decision problem with observations represented as attributed graphs. This approach allows our agents to reason through the lens of relational inductive bias. Agents learn how to reason about hosts interacting with other system entities in a more general manner, and their actions are understood as edits to the graph representing the environment. By introducing this bias, we will show that our agents can better reason about the states of networks and zero-shot adapt to new ones. We show that this approach outperforms the state-of-the-art by a wide margin, and makes our agents capable of defending never-before-seen networks against a wide range of adversaries in a variety of complex, and multi-agent environments.

Inductive learning aims to construct general models from specific examples, guided by biases that influence hypothesis selection and determine generalization capacity. In this work, we focus on characterizing the relational inductive biases present in attention mechanisms, understood as assumptions about the underlying relationships between data elements. From the perspective of geometric deep learning, we analyze the most common attention mechanisms in terms of their equivariance properties with respect to permutation subgroups, which allows us to propose a classification based on their relational biases. Under this perspective, we show that different attention layers are characterized by the underlying relationships they assume on the input data.

Knowledge distillation typically transfers knowledge from a teacher model to a student model by minimizing differences between their output distributions. However, existing distillation approaches largely focus on mimicking absolute probabilities and neglect the valuable relational inductive biases embedded in the teacher's relative predictions, leading to exposure bias. In this paper, we propose Group Relative Knowledge Distillation (GRKD), a novel framework that distills teacher knowledge by learning the relative ranking among classes, rather than directly fitting the absolute distribution. Specifically, we introduce a group relative loss that encourages the student model to preserve the pairwise preference orderings provided by the teacher's outputs. Extensive experiments on classification benchmarks demonstrate that GRKD achieves superior generalization compared to existing methods, especially in tasks requiring fine-grained class differentiation. Our method provides a new perspective on exploiting teacher knowledge, focusing on relational structure rather than absolute likelihood.

This paper explores the impact of relational state abstraction on sample efficiency and performance in collaborative Multi-Agent Reinforcement Learning. The proposed abstraction is based on spatial relationships in environments where direct communication between agents is not allowed, leveraging the ubiquity of spatial reasoning in real-world multi-agent scenarios. We introduce MARC (Multi-Agent Relational Critic), a simple yet effective critic architecture incorporating spatial relational inductive biases by transforming the state into a spatial graph and processing it through a relational graph neural network. The performance of MARC is evaluated across six collaborative tasks, including a novel environment with heterogeneous agents. We conduct a comprehensive empirical analysis, comparing MARC against state-of-the-art MARL baselines, demonstrating improvements in both sample efficiency and asymptotic performance, as well as its potential for generalization. Our findings suggest that a minimal integration of spatial relational inductive biases as abstraction can yield substantial benefits without requiring complex designs or task-specific engineering. This work provides insights into the potential of relational state abstraction to address sample efficiency, a key challenge in MARL, offering a promising direction for developing more efficient algorithms in spatially complex environments.

The human cognitive system exhibits remarkable flexibility and generalization capabilities, partly due to its ability to form low-dimensional, compositional representations of the environment. In contrast, standard neural network architectures often struggle with abstract reasoning tasks, overfitting, and requiring extensive data for training. This paper investigates the impact of the relational bottleneck -- a mechanism that focuses processing on relations among inputs -- on the learning of factorized representations conducive to compositional coding and the attendant flexibility of processing. We demonstrate that such a bottleneck not only improves generalization and learning efficiency, but also aligns network performance with human-like behavioral biases. Networks trained with the relational bottleneck developed orthogonal representations of feature dimensions latent in the dataset, reflecting the factorized structure thought to underlie human cognitive flexibility. Moreover, the relational network mimics human biases towards regularity without pre-specified symbolic primitives, suggesting that the bottleneck fosters the emergence of abstract representations that confer flexibility akin to symbols.

Conditioning image generation on specific features of the desired output is a key ingredient of modern generative models. Most existing approaches focus on conditioning the generation based on free-form text, while some niche studies use scene graphs to describe the content of the image to be generated. This paper explores novel methods to condition image generation that are based on object-centric relational representations. In particular, we propose a methodology to condition the generation of a particular object in an image on the attributed graph representing its structure and associated style. We show that such architectural biases entail properties that facilitate the manipulation and conditioning of the generative process and allow for regularizing the training procedure. The proposed framework is implemented by means of a neural network architecture combining convolutional operators that operate on both the underlying graph and the 2D grid that becomes the output image. The resulting model learns to generate multi-channel masks of the object that can be used as a soft inductive bias in the downstream generative task. Empirical results show that the proposed approach compares favorably against relevant baselines on image generation conditioned on human poses.

Note: This post is a modified excerpt from the introduction to my PhD thesis. Our goal in building machine learning systems is, with rare exceptions, to create algorithms whose utility extends beyond the dataset in which they are trained. In other words, we desire intelligent systems that are capable of generalizing to future data. The process of leveraging observations to draw inferences about the unobserved is the principle of inductionTerminological note: In a non-technical setting, the term inductive – denoting the inference of general laws from particular instances – is typically contrasted with the adjective deductive, which denotes the inference of particular instances from general laws. This broad definition of induction may be used in machine learning to describe, for example, the model fitting process as the inductive step and the deployment on new data as the deductive step. By the same token, some AI methods such as automated theorem provers are described as deductive.

Endowing robots with human-like physical reasoning abilities remains challenging. We argue that existing methods often disregard spatio-temporal relations and by using Graph Neural Networks (GNNs) that incorporate a relational inductive bias, we can shift the learning process towards exploiting relations. In this work, we learn action-conditional forward dynamics models of a simulated manipulation task from visual observations involving cluttered and irregularly shaped objects. We investigate two GNN approaches and empirically assess their capability to generalize to scenarios with novel and an increasing number of objects. The first, Graph Networks (GN) based approach, considers explicitly defined edge attributes and not only does it consistently underperform an auto-encoder baseline that we modified to predict future states, our results indicate how different edge attributes can significantly influence the predictions. Consequently, we develop the Auto-Predictor that does not rely on explicitly defined edge attributes. It outperforms the baseline and the GN-based models. Overall, our results show the sensitivity of GNN-based approaches to the task representation, the efficacy of relational inductive biases and advocate choosing lightweight approaches that implicitly reason about relations over ones that leave these decisions to human designers.

Drug repositioning is an attractive cost-efficient strategy for the development of treatments for human diseases. Here, we propose an interpretable model that learns disease self-representations for drug repositioning. Our self-representation model represents each disease as a linear combination of a few other diseases. We enforce the proximity between diseases to preserve the geometric structure of the human phenome network - a domain-specific knowledge that naturally adds relational inductive bias to the disease self-representations. We prove that our method is globally optimal and show results outperforming state-of-the-art drug repositioning approaches. We further show that the disease self-representations are biologically interpretable.

Many potential applications of reinforcement learning in the real world involve interacting with other agents whose numbers vary over time. We propose new neural policy architectures for these multi-agent problems. In contrast to other methods of training an individual, discrete policy for each agent and then enforcing cooperation through some additional inter-policy mechanism, we follow the spirit of recent work on the power of relational inductive biases in deep networks by learning multi-agent relationships at the policy level via an attentional architecture. In our method, all agents share the same policy, but independently apply it in their own context to aggregate the other agents' state information when selecting their next action. The structure of our architectures allow them to be applied on environments with varying numbers of agents. We demonstrate our architecture on a benchmark multi-agent autonomous vehicle coordination problem, obtaining superior results to a full-knowledge, fully-centralized reference solution, and significantly outperforming it when scaling to large numbers of agents.