Some traits making a "good" AI model are hard to describe upfront. For example, should responses be more polite or more casual? Such traits are sometimes summarized as model character or personality. Without a clear objective, conventional benchmarks based on automatic validation struggle to measure such traits. Evaluation methods using human feedback such as Chatbot Arena have emerged as a popular alternative. These methods infer "better" personality and other desirable traits implicitly by ranking multiple model responses relative to each other. Recent issues with model releases highlight limitations of these existing opaque evaluation approaches: a major model was rolled back over sycophantic personality issues, models were observed overfitting to such feedback-based leaderboards. Despite these known issues, limited public tooling exists to explicitly evaluate model personality. We introduce Feedback Forensics: an open-source toolkit to track AI personality changes, both those encouraged by human (or AI) feedback, and those exhibited across AI models trained and evaluated on such feedback. Leveraging AI annotators, our toolkit enables investigating personality via Python API and browser app. We demonstrate the toolkit's usefulness in two steps: (A) first we analyse the personality traits encouraged in popular human feedback datasets including Chatbot Arena, MultiPref and PRISM; and (B) then use our toolkit to analyse how much popular models exhibit such traits. We release (1) our Feedback Forensics toolkit alongside (2) a web app tracking AI personality in popular models and feedback datasets as well as (3) the underlying annotation data at https://github.com/rdnfn/feedback-forensics.

Reinforcement learning from human feedback (RLHF) is a variant of reinforcement learning (RL) that does not require an engineered reward function but instead learns from human feedback. Due to its increasing popularity, various authors have studied how to learn an accurate reward model from only few samples, making optimal use of this feedback. Because of the cost and complexity of user studies, however, this research is often conducted with synthetic human feedback. Such feedback can be generated by evaluating behavior based on ground-truth rewards which are available for some benchmark tasks. While this setting can help evaluate some aspects of RLHF, it differs from practical settings in which synthetic feedback is not available. Working with real human feedback brings additional challenges that cannot be observed with synthetic feedback, including fatigue, inter-rater inconsistencies, delay, misunderstandings, and modality-dependent difficulties. We describe and discuss some of these challenges together with current practices and opportunities for further research in this paper.

Binary choices, as often used for reinforcement learning from human feedback (RLHF), convey only the *direction* of a preference. A person may choose apples over oranges and bananas over grapes, but *which preference is stronger*? Strength is crucial for decision-making under uncertainty and generalization of preference models, but hard to measure reliably. Metadata such as response times and inter-annotator agreement can serve as proxies for strength, but are often noisy and confounded. We propose ResponseRank to address the challenge of learning from noisy strength signals. Our method uses relative differences in these signals to *rank responses to pairwise comparisons by their inferred preference strength*. Signals are only considered locally within carefully constructed strata, controlling for systemic variation. This enables robust learning of utility differences consistent with strength-derived rankings, all while making minimal assumptions. Our contributions are threefold: (1) ResponseRank, a novel method that robustly learns preference strength by leveraging locally valid relative strength signals; (2) empirical evidence of improved sample efficiency and robustness across diverse tasks: synthetic preference learning (with simulated response times), language modeling (with annotator agreement), and RL control tasks (with simulated episode returns); and (3) the *Pearson Distance Correlation (PDC)*, a novel metric that isolates cardinal utility learning from ordinal accuracy.

Defining a reward function is usually a challenging but critical task for the system designer in reinforcement learning, especially when specifying complex behaviors. Reinforcement learning from human feedback (RLHF) emerges as a promising approach to circumvent this. In RLHF, the agent typically learns a reward function by querying a human teacher using pairwise comparisons of trajectory segments. A key question in this domain is how to reduce the number of queries necessary to learn an informative reward function since asking a human teacher too many queries is impractical and costly. To tackle this question, we propose DUO, a novel method for diverse, uncertain, on-policy query generation and selection in RLHF. Our method produces queries that are (1) more relevant for policy training (via an on-policy criterion), (2) more informative (via a principled measure of epistemic uncertainty), and (3) diverse (via a clustering-based filter). Experimental results on a variety of locomotion and robotic manipulation tasks demonstrate that our method can outperform state-of-the-art RLHF methods given the same total budget of queries, while being robust to possibly irrational teachers.

This paper investigates the remaining useful lifetime (RUL) estimation of bearings under dynamic, i.e., time-varying, operating conditions (OC). Unlike conventional studies that assume constant OC in bearing accelerated life tests, we introduce a dataset with time-varying OC during run-to-failure experiments, simulating real-world scenarios. We explore data-driven approaches to identify the transition point from a healthy to an unhealthy state and estimate the RUL. Additionally, we examine strategies for integrating OC information to enhance RUL estimations. These methodologies are evaluated through numerical experiments using various machine learning algorithms.

In this paper, we advocate for the potential of reinforcement learning from human feedback (RLHF) with self-supervised pretraining to increase the viability of reinforcement learning (RL) for real-world tasks, especially in the context of cyber-physical systems (CPS). We identify potential benefits of self-supervised pretraining in terms of the query sample complexity, safety, robustness, reward exploration and transfer. We believe that exploiting these benefits, combined with the generally improving sample efficiency of RL, will likely enable RL and RLHF to play an increasing role in CPS in the future.

We study the piecewise-stationary dueling bandits problem with $K$ arms, where the time horizon $T$ consists of $M$ stationary segments, each of which is associated with its own preference matrix. The learner repeatedly selects a pair of arms and observes a binary preference between them as feedback. To minimize the accumulated regret, the learner needs to pick the Condorcet winner of each stationary segment as often as possible, despite preference matrices and segment lengths being unknown. We propose the Beat the Winner Reset algorithm and prove a bound on its expected binary weak regret in the stationary case, which tightens the bound of current state-of-art algorithms. We also show a regret bound for the non-stationary case, without requiring knowledge of $M$ or $T$. We further propose and analyze two meta-algorithms, DETECT for weak regret and Monitored Dueling Bandits for strong regret, both based on a detection-window approach that can incorporate any dueling bandit algorithm as a black-box algorithm. Finally, we prove a worst-case lower bound for expected weak regret in the non-stationary case.

Properly defining a reward signal to efficiently train a reinforcement learning (RL) agent is a challenging task. Designing balanced objective functions from which a desired behavior can emerge requires expert knowledge, especially for complex environments. Learning rewards from human feedback or using large language models (LLMs) to directly provide rewards are promising alternatives, allowing non-experts to specify goals for the agent. However, black-box reward models make it difficult to debug the reward. In this work, we propose Object-Centric Assessment with Language Models (OCALM) to derive inherently interpretable reward functions for RL agents from natural language task descriptions. OCALM uses the extensive world-knowledge of LLMs while leveraging the object-centric nature common to many environments to derive reward functions focused on relational concepts, providing RL agents with the ability to derive policies from task descriptions.

Continual learning is a subfield of machine learning, which aims to allow machine learning models to continuously learn on new data, by accumulating knowledge without forgetting what was learned in the past. In this work, we take a step back, and ask: "Why should one care about continual learning in the first place?". We set the stage by examining recent continual learning papers published at four major machine learning conferences, and show that memory-constrained settings dominate the field. Then, we discuss five open problems in machine learning, and even though they might seem unrelated to continual learning at first sight, we show that continual learning will inevitably be part of their solution. These problems are model editing, personalization and specialization, on-device learning, faster (re-)training and reinforcement learning. Finally, by comparing the desiderata from these unsolved problems and the current assumptions in continual learning, we highlight and discuss four future directions for continual learning research. We hope that this work offers an interesting perspective on the future of continual learning, while displaying its potential value and the paths we have to pursue in order to make it successful. This work is the result of the many discussions the authors had at the Dagstuhl seminar on Deep Continual Learning, in March 2023.

In the past couple of years, various approaches to representing and quantifying different types of predictive uncertainty in machine learning, notably in the setting of classification, have been proposed on the basis of second-order probability distributions, i.e., predictions in the form of distributions on probability distributions. A completely conclusive solution has not yet been found, however, as shown by recent criticisms of commonly used uncertainty measures associated with second-order distributions, identifying undesirable theoretical properties of these measures. In light of these criticisms, we propose a set of formal criteria that meaningful uncertainty measures for predictive uncertainty based on second-order distributions should obey. Moreover, we provide a general framework for developing uncertainty measures to account for these criteria, and offer an instantiation based on the Wasserstein distance, for which we prove that all criteria are satisfied.

Reinforcement learning from human feedback (RLHF) is a variant of reinforcement learning (RL) that learns from human feedback instead of relying on an engineered reward function. Building on prior work on the related setting of preference-based reinforcement learning (PbRL), it stands at the intersection of artificial intelligence and human-computer interaction. This positioning provides a promising approach to enhance the performance and adaptability of intelligent systems while also improving the alignment of their objectives with human values. The success in training large language models (LLMs) has impressively demonstrated this potential in recent years, where RLHF has played a decisive role in directing the model's capabilities towards human objectives. This article provides an overview of the fundamentals of RLHF, exploring how RL agents interact with human feedback. While recent focus has been on RLHF for LLMs, our survey covers the technique across multiple domains. We provide our most comprehensive coverage in control and robotics, where many fundamental techniques originate, alongside a dedicated LLM section. We examine the core principles that underpin RLHF, how algorithms and human feedback work together, and the main research trends in the field. Our goal is to give researchers and practitioners a clear understanding of this rapidly growing field.

Algorithm configuration (AC) is concerned with the automated search of the most suitable parameter configuration of a parametrized algorithm. There is currently a wide variety of AC problem variants and methods proposed in the literature. Existing reviews do not take into account all derivatives of the AC problem, nor do they offer a complete classification scheme. To this end, we introduce taxonomies to describe the AC problem and features of configuration methods, respectively. We review existing AC literature within the lens of our taxonomies, outline relevant design choices of configuration approaches, contrast methods and problem variants against each other, and describe the state of AC in industry. Finally, our review provides researchers and practitioners with a look at future research directions in the field of AC.

We consider the combinatorial bandits problem with semi-bandit feedback under finite sampling budget constraints, in which the learner can carry out its action only for a limited number of times specified by an overall budget. The action is to choose a set of arms, whereupon feedback for each arm in the chosen set is received. Unlike existing works, we study this problem in a non-stochastic setting with subset-dependent feedback, i.e., the semi-bandit feedback received could be generated by an oblivious adversary and also might depend on the chosen set of arms. In addition, we consider a general feedback scenario covering both the numerical-based as well as preference-based case and introduce a sound theoretical framework for this setting guaranteeing sensible notions of optimal arms, which a learner seeks to find. We suggest a generic algorithm suitable to cover the full spectrum of conceivable arm elimination strategies from aggressive to conservative. Theoretical questions about the sufficient and necessary budget of the algorithm to find the best arm are answered and complemented by deriving lower bounds for any learning algorithm for this problem scenario.

Uncertainty quantification has received increasing attention in machine learning in the recent past. In particular, a distinction between aleatoric and epistemic uncertainty has been found useful in this regard. The latter refers to the learner's (lack of) knowledge and appears to be especially difficult to measure and quantify. In this paper, we analyse a recent proposal based on the idea of a second-order learner, which yields predictions in the form of distributions over probability distributions. While standard (first-order) learners can be trained to predict accurate probabilities, namely by minimising suitable loss functions on sample data, we show that loss minimisation does not work for second-order predictors: The loss functions proposed for inducing such predictors do not incentivise the learner to represent its epistemic uncertainty in a faithful way.

Heated tool butt welding is a method often used for joining thermoplastics, especially when the components are made out of different materials. The quality of the connection between the components crucially depends on a suitable choice of the parameters of the welding process, such as heating time, temperature, and the precise way how the parts are then welded. Moreover, when different materials are to be joined, the parameter values need to be tailored to the specifics of the respective material. To this end, in this paper, three approaches to tailor the parameter values to optimize the quality of the connection are compared: a heuristic by Potente, statistical experimental design, and Bayesian optimization. With the suitability for practice in mind, a series of experiments are carried out with these approaches, and their capabilities of proposing well-performing parameter values are investigated. As a result, Bayesian optimization is found to yield peak performance, but the costs for optimization are substantial. In contrast, the Potente heuristic does not require any experimentation and recommends parameter values with competitive quality.

The reliable identification of the “best” arm while keeping the sample complexity as low as possible is a common task in the field of multi-armed bandits. In the multi-dueling variant of multi-armed bandits, where feedback is provided in the form of a winning arm among as set of k chosen ones, a reasonable notion of best arm is the generalized Condorcet winner (GCW). The latter is an the arm that has the greatest probability of being the winner in each subset containing it. In this paper, we derive lower bounds on the sample complexity for the task of identifying the GCW under various assumptions. As a by-product, our lower bound results provide new insights for the special case of dueling bandits (k = 2). We propose the Dvoretzky–Kiefer–Wolfowitz tournament (DKWT) algorithm, which we prove to be nearly optimal. In a numerical study, we show that DKWT empirically outperforms current state-of-the-art algorithms, even in the special case of dueling bandits or under a Plackett-Luce assumption on the feedback mechanism.

The problem of reaching a consensus, i.e., driving the state of systems interacting in a network to a common value, has many interesting applications and has recently received much attention within the control community. Consensus is reached through a local interaction rule between the interconnected systems that may require a synchronous updating, or an asynchronous updating based on gossip or broadcast communications. The objective of this talk is to review some consensus based approaches recently proposed by the Automatic Control group of the University of Cagliari. These approaches address different problems in a distributed setting, such as such as estimation of network properties, averaging, and load balancing. The emerging behavior of a multi-agent system is strictly related to the topology of the underlying network, which in turn is related to the spectrum of the Laplacian matrix of the network graph. In the first part of the talk, a decentralized algorithm to estimate the Laplacian eigenvalues is presented. The basic idea is to provide a local interaction rule among agents so that their state trajectory is a linear combination of sinusoids oscillating only at frequencies corresponding to the eigenvalues of the Laplacian matrix. In this way, the problem of decentralized estimation of the eigenvalues is mapped into a standard signal processing problem. In the second part of the talk two extensions of the classical consensus algorithm are presented. One extension concerns discrete consensus under gossip communications. In quantized consensus the state of the agents takes discrete values. This setting can be extended to encompass the more general problem of load balancing of discrete tasks of different size over a set of heterogeneous processors. To solve this problem the notion of swap, i.e., an exchange of tasks between agent that does not balance the load, is fundamental to avoid local minima of the objective function. A new deterministic swapping rule that- greatly reduces the expected time to reach consensus will be discussed. Another extension concerns a new decentralized discontinuous interaction rule which allows to synchronize a network of agents with first-order dynamics. The topology of the network is described by a strongly connected directed graph, and the agents dynamics are perturbed by additive and uniformly bounded uncertain timevarying disturbances. The proposed interaction rule completely rejects the effects of the disturbances by providing the achievement of consensus after a finite transient time which can be arbitrarily reduced by increasing a scalar control parameter.