This paper aims at i) formally understanding equilibria reached by such agents, and ii) matching emergent phenomena of such equilibria to real-world targets. Parameter sharing with decentralized execution has been introduced as an efficient way to train multiple agents using a single policy network.

Summary and Contributions: The paper presents the concept of shared equilibrium in certain kinds of multi agent stochastic games with a restricted form of partial observability. The formalism includes the notion of supertypes (different distributions of agents) and types (where each agents is given a true type each episode). The agent's type influences the rewards available as does the joint state of the system and joint action over all agents. One key constraint is that all agents of the same type follow the same policy from an egocentric perspective (where they themselves are the focal agent and all other agents are interchangeable). They define a policy gradient approach for individual agents, also present a higher order learning rule that shifts the distribution over supertypes at a slower timescale.

We study a game between liquidity provider and liquidity taker agents interacting in an over-the-counter market, for which the typical example is foreign exchange. We show how a suitable design of parameterized families of reward functions coupled with shared policy learning constitutes an efficient solution to this problem. By playing against each other, our deep-reinforcement-learning-driven agents learn emergent behaviors relative to a wide spectrum of objectives encompassing profit-and-loss, optimal execution and market share. In particular, we find that liquidity providers naturally learn to balance hedging and skewing, where skewing refers to setting their buy and sell prices asymmetrically as a function of their inventory. We further introduce a novel RL-based calibration algorithm which we found performed well at imposing constraints on the game equilibrium. On the theoretical side, we are able to show convergence rates for our multi-agent policy gradient algorithm under a transitivity assumption, closely related to generalized ordinal potential games.

Agent based modelling (ABM) is a computational approach to modelling complex systems by specifying the behaviour of autonomous decision-making components or agents in the system and allowing the system dynamics to emerge from their interactions. Recent advances in the field of Multi-agent reinforcement learning (MARL) have made it feasible to study the equilibrium of complex environments where multiple agents learn simultaneously. However, most ABM frameworks are not RL-native, in that they do not offer concepts and interfaces that are compatible with the use of MARL to learn agent behaviours. In this paper, we introduce a new open-source framework, Phantom, to bridge the gap between ABM and MARL. Phantom is an RL-driven framework for agent-based modelling of complex multi-agent systems including, but not limited to economic systems and markets. The framework aims to provide the tools to simplify the ABM specification in a MARL-compatible way - including features to encode dynamic partial observability, agent utility functions, heterogeneity in agent preferences or types, and constraints on the order in which agents can act (e.g. Stackelberg games, or more complex turn-taking environments). In this paper, we present these features, their design rationale and present two new environments leveraging the framework.

Neural-based word embeddings using solely distributional information have consistently produced useful meaning representations for downstream tasks. However, existing approaches often result in representations that are hard to interpret and control. Natural language definitions, on the other side, possess a recursive, self-explanatory semantic structure that can support novel representation learning paradigms able to preserve explicit conceptual relations and constraints in the vector space. This paper proposes a neuro-symbolic, multi-relational framework to learn word embeddings exclusively from natural language definitions by jointly mapping defined and defining terms along with their corresponding semantic relations. By automatically extracting the relations from definitions corpora and formalising the learning problem via a translational objective, we specialise the framework in hyperbolic space to capture the hierarchical and multi-resolution structure induced by the definitions. An extensive empirical analysis demonstrates that the framework can help impose the desired structural constraints while preserving the mapping required for controllable and interpretable semantic navigation. Moreover, the experiments reveal the superiority of the hyperbolic word embeddings over the euclidean counterparts and demonstrate that the multi-relational framework can obtain competitive results when compared to state-of-the-art neural approaches (including Transformers), with the advantage of being significantly more efficient and intrinsically interpretable.

To be robust to surprising developments, an intelligent agent must be able to respond to many different types of unexpected change in the world. To date, there are no general frameworks for defining and characterizing the types of environment changes that are possible. We introduce a formal and theoretical framework for defining and categorizing environment transformations, changes to the world an agent inhabits. We introduce two types of environment transformation: R-transformations which modify environment dynamics and T-transformations which modify the generation process that produces scenarios. We present a new language for describing domains, scenario generators, and transformations, called the Transformation and Simulator Abstraction Language (T-SAL), and a logical formalism that rigorously defines these concepts. Then, we offer the first formal and computational set of tests for eight categories of environment transformations. This domain-independent framework paves the way for describing unambiguous classes of novelty, constrained and domain-independent random generation of environment transformations, replication of environment transformation studies, and fair evaluation of agent robustness.

Disentangling the encodings of neural models is a fundamental aspect for improving interpretability, semantic control and downstream task performance in Natural Language Processing. Currently, most disentanglement methods are unsupervised or rely on synthetic datasets with known generative factors. We argue that recurrent syntactic and semantic regularities in textual data can be used to provide the models with both structural biases and generative factors. We leverage the semantic structures present in a representative and semantically dense category of sentence types, definitional sentences, for training a Variational Autoencoder to learn disentangled representations. Our experimental results show that the proposed model outperforms unsupervised baselines on several qualitative and quantitative benchmarks for disentanglement, and it also improves the results in the downstream task of definition modeling.

Taxonomies hierarchically organize concepts in a domain. Building and maintaining them by hand is a tedious and time-consuming task. This paper proposes a novel, unsupervised algorithm for automatically learning an IS-A taxonomy from scratch by analyzing a given text corpus. Our approach is designed to deal with infrequently occurring concepts, so it can effectively induce taxonomies even from small corpora. Algorithmically, the approach makes two important contributions. First, it performs inference based on clustering and the distributional semantics, which can capture links among concepts never mentioned together. Second, it uses a novel graph-based algorithm to detect and remove incorrect is-a relations from a taxonomy. An empirical evaluation on five corpora demonstrates the utility of our proposed approach.

We attempt to set a mathematical foundation of immunology and amino acid chains. To measure the similarities of these chains, a kernel on strings is defined using only the sequence of the chains and a good amino acid substitution matrix (e.g. BLOSUM62). The kernel is used in learning machines to predict binding affinities of peptides to human leukocyte antigens DR (HLA-DR) molecules. On both fixed allele (Nielsen and Lund 2009) and pan-allele (Nielsen et.al. 2010) benchmark databases, our algorithm achieves the state-of-the-art performance. The kernel is also used to define a distance on an HLA-DR allele set based on which a clustering analysis precisely recovers the serotype classifications assigned by WHO (Nielsen and Lund 2009, and Marsh et.al. 2010). These results suggest that our kernel relates well the chain structure of both peptides and HLA-DR molecules to their biological functions, and that it offers a simple, powerful and promising methodology to immunology and amino acid chain studies.