We study convergence rates of Hamiltonian Monte Carlo (HMC) algorithms with leapfrog integration under mild conditions on stochastic gradient oracle for the target distribution (SGHMC). Our method extends standard HMC by allowing the use of general auxiliary distributions, which is achieved by a novel procedure of Alternating Directions. The convergence analysis is based on the investigations of the Dirichlet forms associated with the underlying Markov chain driving the algorithms. For this purpose, we provide a detailed analysis on the error of the leapfrog integrator for Hamiltonian motions with both the kinetic and potential energy functions in general form. We characterize the explicit dependence of the convergence rates on key parameters such as the problem dimension, functional properties of both the target and auxiliary distributions, and the quality of the oracle.

Aligning generative models with human preference via RLHF typically suffers from overoptimization, where an imperfectly learned reward model can misguide the generative model to output undesired responses. We investigate this problem in a principled manner by identifying the source of the misalignment as a form of distributional shift and uncertainty in learning human preferences. To mitigate overoptimization, we first propose a theoretical algorithm that chooses the best policy for an adversarially chosen reward model; one that simultaneously minimizes the maximum likelihood estimation of the loss and a reward penalty term. Here, the reward penalty term is introduced to prevent the policy from choosing actions with spurious high proxy rewards, resulting in provable sample efficiency of the algorithm under a partial coverage style condition. Moving from theory to practice, the proposed algorithm further enjoys an equivalent but surprisingly easy-to-implement reformulation. Using the equivalence between reward models and the corresponding optimal policy, the algorithm features a simple objective that combines: (i) a preference optimization loss that directly aligns the policy with human preference, and (ii) a supervised learning loss that explicitly imitates the policy with a (suitable) baseline distribution. In the context of aligning large language models (LLM), this objective fuses the direct preference optimization (DPO) loss with the supervised fune-tuning (SFT) loss to help mitigate the overoptimization towards undesired responses, for which we name the algorithm Regularized Preference Optimization (RPO). Experiments of aligning LLMs demonstrate the improved performance of RPO compared with DPO baselines. Our work sheds light on the interplay between preference optimization and SFT in tuning LLMs with both theoretical guarantees and empirical evidence.

One of the inherent challenges in deploying transformers on time series is that \emph{reality only happens once}; namely, one typically only has access to a single trajectory of the data-generating process comprised of non-i.i.d. observations. We derive non-asymptotic statistical guarantees in this setting through bounds on the \textit{generalization} of a transformer network at a future-time $t$, given that it has been trained using $N\le t$ observations from a single perturbed trajectory of a Markov process. Under the assumption that the Markov process satisfies a log-Sobolev inequality, we obtain a generalization bound which effectively converges at the rate of ${O}(1/\sqrt{N})$. Our bound depends explicitly on the activation function ($\operatorname{Swish}$, $\operatorname{GeLU}$, or $\tanh$ are considered), the number of self-attention heads, depth, width, and norm-bounds defining the transformer architecture. Our bound consists of three components: (I) The first quantifies the gap between the stationary distribution of the data-generating Markov process and its distribution at time $t$, this term converges exponentially to $0$. (II) The next term encodes the complexity of the transformer model and, given enough time, eventually converges to $0$ at the rate ${O}(\log(N)^r/\sqrt{N})$ for any $r>0$. (III) The third term guarantees that the bound holds with probability at least $1$-$\delta$, and converges at a rate of ${O}(\sqrt{\log(1/\delta)}/\sqrt{N})$.

We consider offline reinforcement learning (RL) in $H$-horizon Markov decision processes (MDPs) under the linear $q^\pi$-realizability assumption, where the action-value function of every policy is linear with respect to a given $d$-dimensional feature function. The hope in this setting is that learning a good policy will be possible without requiring a sample size that scales with the number of states in the MDP. Foster et al. [2021] have shown this to be impossible even under $\textit{concentrability}$, a data coverage assumption where a coefficient $C_\text{conc}$ bounds the extent to which the state-action distribution of any policy can veer off the data distribution. However, the data in this previous work was in the form of a sequence of individual transitions. This leaves open the question of whether the negative result mentioned could be overcome if the data was composed of sequences of full trajectories. In this work we answer this question positively by proving that with trajectory data, a dataset of size $\text{poly}(d,H,C_\text{conc})/\epsilon^2$ is sufficient for deriving an $\epsilon$-optimal policy, regardless of the size of the state space. The main tool that makes this result possible is due to Weisz et al. [2023], who demonstrate that linear MDPs can be used to approximate linearly $q^\pi$-realizable MDPs. The connection to trajectory data is that the linear MDP approximation relies on "skipping" over certain states. The associated estimation problems are thus easy when working with trajectory data, while they remain nontrivial when working with individual transitions. The question of computational efficiency under our assumptions remains open.

The purpose of class distribution estimation (also known as quantification) is to determine the values of the prior class probabilities in a test dataset without class label observations. A variety of methods to achieve this have been proposed in the literature, most of them based on the assumption that the distributions of the training and test data are related through prior probability shift (also known as label shift). Among these methods, Friedman's method has recently been found to perform relatively well both for binary and multi-class quantification. We discuss the properties of Friedman's method and another approach mentioned by Friedman (called DeBias method in the literature) in the context of a general framework for designing linear equation systems for class distribution estimation.

Machine learning (ML) provides algorithms to create computer programs based on data without explicitly programming them. In business process management (BPM), ML applications are used to analyse and improve processes efficiently. Three frequent examples of using ML are providing decision support through predictions, discovering accurate process models, and improving resource allocation. This paper organises the body of knowledge on ML in BPM. We extract BPM tasks from different literature streams, summarise them under the phases of a process`s lifecycle, explain how ML helps perform these tasks and identify technical commonalities in ML implementations across tasks. This study is the first exhaustive review of how ML has been used in BPM. We hope that it can open the door for a new era of cumulative research by helping researchers to identify relevant preliminary work and then combine and further develop existing approaches in a focused fashion. Our paper helps managers and consultants to find ML applications that are relevant in the current project phase of a BPM initiative, like redesigning a business process. We also offer - as a synthesis of our review - a research agenda that spreads ten avenues for future research, including applying novel ML concepts like federated learning, addressing less regarded BPM lifecycle phases like process identification, and delivering ML applications with a focus on end-users.

Skills are effective temporal abstractions established for sequential decision making tasks, which enable efficient hierarchical learning for long-horizon tasks and facilitate multi-task learning through their transferability. Despite extensive research, research gaps remain in multi-agent scenarios, particularly for automatically extracting subgroup coordination patterns in a multi-agent task. In this case, we propose two novel auto-encoder schemes: VO-MASD-3D and VO-MASD-Hier, to simultaneously capture subgroup- and temporal-level abstractions and form multi-agent skills, which firstly solves the aforementioned challenge. An essential algorithm component of these schemes is a dynamic grouping function that can automatically detect latent subgroups based on agent interactions in a task. Notably, our method can be applied to offline multi-task data, and the discovered subgroup skills can be transferred across relevant tasks without retraining. Empirical evaluations on StarCraft tasks indicate that our approach significantly outperforms existing methods regarding applying skills in multi-agent reinforcement learning (MARL). Moreover, skills discovered using our method can effectively reduce the learning difficulty in MARL scenarios with delayed and sparse reward signals.

Large Language Models (LLM) based agents have shown promise in autonomously completing tasks across various domains, e.g., robotics, games, and web navigation. However, these agents typically require elaborate design and expert prompts to solve tasks in specific domains, which limits their adaptability. We introduce AutoManual, a framework enabling LLM agents to autonomously build their understanding through interaction and adapt to new environments. AutoManual categorizes environmental knowledge into diverse rules and optimizes them in an online fashion by two agents: 1) The Planner codes actionable plans based on current rules for interacting with the environment. 2) The Builder updates the rules through a well-structured rule system that facilitates online rule management and essential detail retention. To mitigate hallucinations in managing rules, we introduce \textit{case-conditioned prompting} strategy for the Builder. Finally, the Formulator agent compiles these rules into a comprehensive manual. The self-generated manual can not only improve the adaptability but also guide the planning of smaller LLMs while being human-readable. Given only one simple demonstration, AutoManual significantly improves task success rates, achieving 97.4\% with GPT-4-turbo and 86.2\% with GPT-3.5-turbo on ALFWorld benchmark tasks. The source code will be available soon.

Tensor Attention, a multi-view attention that is able to capture high-order correlations among multiple modalities, can overcome the representational limitations of classical matrix attention. However, the $\Omega(n^3)$ time complexity of tensor attention poses a significant obstacle to its practical implementation in transformers, where $n$ is the input sequence length. In this work, we prove that the backward gradient of tensor attention training can be computed in almost linear $n^{1+o(1)}$ time, the same complexity as its forward computation under a bounded entries assumption. We provide a closed-form solution for the gradient and propose a fast computation method utilizing polynomial approximation methods and tensor algebraic tricks. Furthermore, we prove the necessity and tightness of our assumption through hardness analysis, showing that slightly weakening it renders the gradient problem unsolvable in truly subcubic time. Our theoretical results establish the feasibility of efficient higher-order transformer training and may facilitate practical applications of tensor attention architectures.

Causal discovery amounts to unearthing causal relationships amongst features in data. It is a crucial companion to causal inference, necessary to build scientific knowledge without resorting to expensive or impossible randomised control trials. In this paper, we explore how reasoning with symbolic representations can support causal discovery. Specifically, we deploy assumption-based argumentation (ABA), a well-established and powerful knowledge representation formalism, in combination with causality theories, to learn graphs which reflect causal dependencies in the data. We prove that our method exhibits desirable properties, notably that, under natural conditions, it can retrieve ground-truth causal graphs. We also conduct experiments with an implementation of our method in answer set programming (ASP) on four datasets from standard benchmarks in causal discovery, showing that our method compares well against established baselines.