Curriculum learning has been a quiet yet crucial component of many of the high-profile successes of reinforcement learning. Despite this, none of the major reinforcement learning libraries directly support curriculum learning or include curriculum learning implementations. These methods can improve the capabilities and robustness of RL agents, but often require significant, complex changes to agent training code. We introduce Syllabus, a library for training RL agents with curriculum learning, as a solution to this problem. Syllabus provides a universal API for curriculum learning algorithms, implementations of popular curriculum learning methods, and infrastructure for easily integrating them with distributed training code written in nearly any RL library. Syllabus provides a minimal API for each of the core components of curriculum learning, dramatically simplifying the process of designing new algorithms and applying existing algorithms to new environments. We demonstrate that the same Syllabus code can be used to train agents written in multiple different RL libraries on numerous domains. In doing so, we present the first examples of curriculum learning in NetHack and Neural MMO, two of the premier challenges for single-agent and multi-agent RL respectively, achieving strong results compared to state of the art baselines.

LLMs are increasingly fine-tuned using RLHF datasets to align them with human preferences and values. However, very limited research has investigated which specific human values are operationalized through these datasets. In this paper, we introduce Value Imprint, a framework for auditing and classifying the human values embedded within RLHF datasets. To investigate the viability of this framework, we conducted three case study experiments by auditing the Anthropic/hh-rlhf, OpenAI WebGPT Comparisons, and Alpaca GPT-4-LLM datasets to examine the human values embedded within them. Our analysis involved a two-phase process. During the first phase, we developed a taxonomy of human values through an integrated review of prior works from philosophy, axiology, and ethics. Then, we applied this taxonomy to annotate 6,501 RLHF preferences. During the second phase, we employed the labels generated from the annotation as ground truth data for training a transformer-based machine learning model to audit and classify the three RLHF datasets. Through this approach, we discovered that information-utility values, including Wisdom/Knowledge and Information Seeking, were the most dominant human values within all three RLHF datasets. In contrast, prosocial and democratic values, including Well-being, Justice, and Human/Animal Rights, were the least represented human values. These findings have significant implications for developing language models that align with societal values and norms. We contribute our datasets to support further research in this area.

Virtual Reality (VR) has been a beneficial training tool in fields such as advanced manufacturing. However, users may experience a high cognitive load due to various factors, such as the use of VR hardware or tasks within the VR environment. Studies have shown that eye-tracking has the potential to detect cognitive load, but in the context of VR and complex spatiotemporal tasks (e.g., assembly and disassembly), it remains relatively unexplored. Here, we present an ongoing study to detect users' cognitive load using an eye-tracking-based machine learning approach. We developed a VR training system for cold spray and tested it with 22 participants, obtaining 19 valid eye-tracking datasets and NASA-TLX scores. We applied Multi-Layer Perceptron (MLP) and Random Forest (RF) models to compare the accuracy of predicting cognitive load (i.e., NASA-TLX) using pupil dilation and fixation duration. Our preliminary analysis demonstrates the feasibility of using eye tracking to detect cognitive load in complex spatiotemporal VR experiences and motivates further exploration.

Massive Open Online Courses (MOOCs) have greatly contributed to making education more accessible. However, many MOOCs maintain a rigid, one-size-fits-all structure that fails to address the diverse needs and backgrounds of individual learners. Learning path personalization aims to address this limitation, by tailoring sequences of educational content to optimize individual student learning outcomes. Existing approaches, however, often require either massive student interaction data or extensive expert annotation, limiting their broad application. In this study, we introduce a novel data-efficient framework for learning path personalization that operates without expert annotation. Our method employs a flexible recommender system pre-trained with reinforcement learning on a dataset of raw course materials. Through experiments on semi-synthetic data, we show that this pre-training stage substantially improves data-efficiency in a range of adaptive learning scenarios featuring new educational materials. This opens up new perspectives for the design of foundation models for adaptive learning.

Federated learning (FL) is an emerging paradigm for training machine learning models across distributed clients. Traditionally, in FL settings, a central server assigns training efforts (or strategies) to clients. However, from a market-oriented perspective, clients may independently choose their training efforts based on rational self-interest. To explore this, we propose a potential game framework where each client's payoff is determined by their individual efforts and the rewards provided by the server. The rewards are influenced by the collective efforts of all clients and can be modulated through a reward factor. Our study begins by establishing the existence of Nash equilibria (NEs), followed by an investigation of uniqueness in homogeneous settings. We demonstrate a significant improvement in clients' training efforts at a critical reward factor, identifying it as the optimal choice for the server. Furthermore, we prove the convergence of the best-response algorithm to compute NEs for our FL game. Finally, we apply the training efforts derived from specific NEs to a real-world FL scenario, validating the effectiveness of the identified optimal reward factor.

Natural policy gradient (NPG) is a common policy optimization algorithm and can be viewed as mirror ascent in the space of probabilities. Recently, Vaswani et al. [2021] introduced a policy gradient method that corresponds to mirror ascent in the dual space of logits. We refine this algorithm, removing its need for a normalization across actions and analyze the resulting method (referred to as SPMA). For tabular MDPs, we prove that SPMA with a constant step-size matches the linear convergence of NPG and achieves a faster convergence than constant step-size (accelerated) softmax policy gradient. To handle large state-action spaces, we extend SPMA to use a log-linear policy parameterization. Unlike that for NPG, generalizing SPMA to the linear function approximation (FA) setting does not require compatible function approximation. Unlike MDPO, a practical generalization of NPG, SPMA with linear FA only requires solving convex softmax classification problems. We prove that SPMA achieves linear convergence to the neighbourhood of the optimal value function. We extend SPMA to handle non-linear FA and evaluate its empirical performance on the MuJoCo and Atari benchmarks. Our results demonstrate that SPMA consistently achieves similar or better performance compared to MDPO, PPO and TRPO.

We propose a new approach for fine-grained uncertainty quantification (UQ) using a collision matrix. For a classification problem involving $K$ classes, the $K\times K$ collision matrix $S$ measures the inherent (aleatoric) difficulty in distinguishing between each pair of classes. In contrast to existing UQ methods, the collision matrix gives a much more detailed picture of the difficulty of classification. We discuss several possible downstream applications of the collision matrix, establish its fundamental mathematical properties, as well as show its relationship with existing UQ methods, including the Bayes error rate. We also address the new problem of estimating the collision matrix using one-hot labeled data. We propose a series of innovative techniques to estimate $S$. First, we learn a contrastive binary classifier which takes two inputs and determines if they belong to the same class. We then show that this contrastive classifier (which is PAC learnable) can be used to reliably estimate the Gramian matrix of $S$, defined as $G=S^TS$. Finally, we show that under very mild assumptions, $G$ can be used to uniquely recover $S$, a new result on stochastic matrices which could be of independent interest. Experimental results are also presented to validate our methods on several datasets.

This book is about conformal prediction and related inferential techniques that build on permutation tests and exchangeability. These techniques are useful in a diverse array of tasks, including hypothesis testing and providing uncertainty quantification guarantees for machine learning systems. Much of the current interest in conformal prediction is due to its ability to integrate into complex machine learning workflows, solving the problem of forming prediction sets without any assumptions on the form of the data generating distribution. Since contemporary machine learning algorithms have generally proven difficult to analyze directly, conformal prediction's main appeal is its ability to provide formal, finite-sample guarantees when paired with such methods. The goal of this book is to teach the reader about the fundamental technical arguments that arise when researching conformal prediction and related questions in distribution-free inference. Many of these proof strategies, especially the more recent ones, are scattered among research papers, making it difficult for researchers to understand where to look, which results are important, and how exactly the proofs work. We hope to bridge this gap by curating what we believe to be some of the most important results in the literature and presenting their proofs in a unified language, with illustrations, and with an eye towards pedagogy.

Matching markets have become increasingly relevant in a variety of modern applications, including but not limited to school admissions, organ transplantation, and job matching. Traditionally, these markets were studied under the assumption that the demand side agents (aka players or agents) and the supply side agents (aka arms) have fixed, known preferences, allowing for stable matching via the deferred acceptance algorithm like the Gale-Shapley algorithm introduced in Gale and Shapley (1962). However, in applications like crowdsourcing, online labor markets, and finance, the preferences are not given to the agents, and they must learn it over time by interacting with the environment. Modeling the matching market as multi-agent, multi-armed competitive bandits, there has been extensive work on various aspects, including coordinated centralized matching, decentralized matching, and game-theoretic analysis (see Liu et al. (2020); Basu et al. (2021); Sankararaman et al. (2021); Etesami and Srikant (2024)). In this paper, we study a large and dynamic matching market where the number of arms K is large and often exceeds the number of agents N( K).

We propose a unifying framework for methods that perform Bayesian online learning in non-stationary environments. We call the framework BONE, which stands for (B)ayesian (O)nline learning in (N)on-stationary (E)nvironments. BONE provides a common structure to tackle a variety of problems, including online continual learning, prequential forecasting, and contextual bandits. The framework requires specifying three modelling choices: (i) a model for measurements (e.g., a neural network), (ii) an auxiliary process to model non-stationarity (e.g., the time since the last changepoint), and (iii) a conditional prior over model parameters (e.g., a multivariate Gaussian). The framework also requires two algorithmic choices, which we use to carry out approximate inference under this framework: (i) an algorithm to estimate beliefs (posterior distribution) about the model parameters given the auxiliary variable, and (ii) an algorithm to estimate beliefs about the auxiliary variable. We show how this modularity allows us to write many different existing methods as instances of BONE; we also use this framework to propose a new method. We then experimentally compare existing methods with our proposed new method on several datasets; we provide insights into the situations that make one method more suitable than another for a given task.