In this work, we study probability functions associated with Gaussian mixture models. Our primary focus is on extending the use of spherical radial decomposition for multivariate Gaussian random vectors to the context of Gaussian mixture models, which are not inherently spherical but only conditionally so. Specifically, the conditional probability distribution, given a random parameter of the random vector, follows a Gaussian distribution, allowing us to apply Bayesian analysis tools to the probability function. This assumption, together with spherical radial decomposition for Gaussian random vectors, enables us to represent the probability function as an integral over the Euclidean sphere. Using this representation, we establish sufficient conditions to ensure the differentiability of the probability function and provide and integral representation of its gradient. Furthermore, leveraging the Bayesian decomposition, we approximate the probability function using random sampling over the parameter space and the Euclidean sphere. Finally, we present numerical examples that illustrate the advantages of this approach over classical approximations based on random vector sampling.

Point processes are widely used statistical models for uncovering the temporal patterns in dependent event data. In many applications, the event time cannot be observed exactly, calling for the incorporation of time uncertainty into the modeling of point process data. In this work, we introduce a framework to model time-uncertain point processes possibly on a network. We start by deriving the formulation in the continuous-time setting under a few assumptions motivated by application scenarios. After imposing a time grid, we obtain a discrete-time model that facilitates inference and can be computed by first-order optimization methods such as Gradient Descent or Variation inequality (VI) using batch-based Stochastic Gradient Descent (SGD). The parameter recovery guarantee is proved for VI inference at an $O(1/k)$ convergence rate using $k$ SGD steps. Our framework handles non-stationary processes by modeling the inference kernel as a matrix (or tensor on a network) and it covers the stationary process, such as the classical Hawkes process, as a special case. We experimentally show that the proposed approach outperforms previous General Linear model (GLM) baselines on simulated and real data and reveals meaningful causal relations on a Sepsis-associated Derangements dataset.

General state-space models (SSMs) are widely used in statistical machine learning and are among the most classical generative models for sequential time-series data. SSMs, comprising latent Markovian states, can be subjected to variational inference (VI), but standard VI methods like the importance-weighted autoencoder (IWAE) lack functionality for streaming data. To enable online VI in SSMs when the observations are received in real time, we propose maximising an IWAE-type variational lower bound on the asymptotic contrast function, rather than the standard IWAE ELBO, using stochastic approximation. Unlike the recursive maximum likelihood method, which directly maximises the asymptotic contrast, our approach, called online sequential IWAE (OSIWAE), allows for online learning of both model parameters and a Markovian recognition model for inferring latent states. By approximating filter state posteriors and their derivatives using sequential Monte Carlo (SMC) methods, we create a particle-based framework for online VI in SSMs. This approach is more theoretically well-founded than recently proposed online variational SMC methods. We provide rigorous theoretical results on the learning objective and a numerical study demonstrating the method's efficiency in learning model parameters and particle proposal kernels.

Efficient sampling from un-normalized target distributions is pivotal in scientific computing and machine learning. While neural samplers have demonstrated potential with a special emphasis on sampling efficiency, existing neural implicit samplers still have issues such as poor mode covering behavior, unstable training dynamics, and sub-optimal performances. To tackle these issues, in this paper, we introduce Denoising Fisher Training (DFT), a novel training approach for neural implicit samplers with theoretical guarantees. We frame the training problem as an objective of minimizing the Fisher divergence by deriving a tractable yet equivalent loss function, which marks a unique theoretical contribution to assessing the intractable Fisher divergences. DFT is empirically validated across diverse sampling benchmarks, including two-dimensional synthetic distribution, Bayesian logistic regression, and high-dimensional energy-based models (EBMs). Notably, in experiments with high-dimensional EBMs, our best one-step DFT neural sampler achieves results on par with MCMC methods with up to 200 sampling steps, leading to a substantially greater efficiency over 100 times higher. This result not only demonstrates the superior performance of DFT in handling complex high-dimensional sampling but also sheds light on efficient sampling methodologies across broader applications.

This paper develops theory and algorithms for interacting large language model agents (LLMAs) using methods from statistical signal processing and microeconomics. While both fields are mature, their application to decision-making by interacting LLMAs remains unexplored. Motivated by Bayesian sentiment analysis on online platforms, we construct interpretable models and stochastic control algorithms that enable LLMAs to interact and perform Bayesian inference. Because interacting LLMAs learn from prior decisions and external inputs, they exhibit bias and herding behavior. Thus, developing interpretable models and stochastic control algorithms is essential to understand and mitigate these behaviors. This paper has three main results. First, we show using Bayesian revealed preferences from microeconomics that an individual LLMA satisfies the sufficient conditions for rationally inattentive (bounded rationality) utility maximization and, given an observation, the LLMA chooses an action that maximizes a regularized utility. Second, we utilize Bayesian social learning to construct interpretable models for LLMAs that interact sequentially with each other and the environment while performing Bayesian inference. Our models capture the herding behavior exhibited by interacting LLMAs. Third, we propose a stochastic control framework to delay herding and improve state estimation accuracy under two settings: (a) centrally controlled LLMAs and (b) autonomous LLMAs with incentives. Throughout the paper, we demonstrate the efficacy of our methods on real datasets for hate speech classification and product quality assessment, using open-source models like Mistral and closed-source models like ChatGPT. The main takeaway of this paper, based on substantial empirical analysis and mathematical formalism, is that LLMAs act as rationally bounded Bayesian agents that exhibit social learning when interacting.

In-context learning (ICL) is a powerful technique for getting language models to perform complex tasks with no training updates. Prior work has established strong correlations between the number of in-context examples provided and the accuracy of the model's predictions. In this paper, we seek to explain this correlation by showing that ICL approximates a Bayesian learner. This perspective gives rise to a family of novel Bayesian scaling laws for ICL. In experiments with \mbox{GPT-2} models of different sizes, our scaling laws exceed or match existing scaling laws in accuracy while also offering interpretable terms for task priors, learning efficiency, and per-example probabilities. To illustrate the analytic power that such interpretable scaling laws provide, we report on controlled synthetic dataset experiments designed to inform real-world studies of safety alignment. In our experimental protocol, we use SFT to suppress an unwanted existing model capability and then use ICL to try to bring that capability back (many-shot jailbreaking). We then experiment on real-world instruction-tuned LLMs using capabilities benchmarks as well as a new many-shot jailbreaking dataset. In all cases, Bayesian scaling laws accurately predict the conditions under which ICL will cause the suppressed behavior to reemerge, which sheds light on the ineffectiveness of post-training at increasing LLM safety.

Hawkes processes are a class of past-dependent point processes, widely used in many applications such as seismology [Ogata, 1988], criminology [Olinde and Short, 2020] and neuroscience [Reynaud-Bouret et al., 2013] for their ability to capture complex dependence structures. In their multidimensional version [Ogata, 1988], Hawkes processes can model pairwise interactions between different types of events, allowing to recover a connectivity graph between different features. Originally developed by Hawkes [1971] in order to model self-exciting phenomena, where each event increases the probability of a new event occurring, many extensions have been proposed ever since. In particular, nonlinear Hawkes processes have been introduced notably to detect inhibiting interactions, when an event can decrease the probability of another one appearing. Hawkes processes with inhibition are notoriously more complicated to handle due to the loss of many properties of linear Hawkes processes such as the cluster representation and the branching structure of the process [Hawkes and Oakes, 1974]. Since the first article on nonlinear Hawkes processes [Brémaud and Massoulié, 1996] proving in particular their existence, many works have focused on inhibition in the past few years. Among them, limit theorems have been established in [Costa et al., 2020] while Duval et al. [2022] obtained mean-field results on the behaviour of two neuronal populations. Regarding statistical inference, in the frequentist setting we can mention the exact maximum likelihood procedure of Bonnet et al. [2023], the least-squares approach by Bacry et al. [2020] and the nonparametric approach based on Bernstein-type polynomials by Lemonnier and Vayatis [2014]. While the first one proposes an exact inference procedure, it is restricted to exponential kernels.

Lifelong reinforcement learning (RL) has been developed as a paradigm for extending single-task RL to more realistic, dynamic settings. In lifelong RL, the "life" of an RL agent is modeled as a stream of tasks drawn from a task distribution. We propose EPIC (\underline{E}mpirical \underline{P}AC-Bayes that \underline{I}mproves \underline{C}ontinuously), a novel algorithm designed for lifelong RL using PAC-Bayes theory. EPIC learns a shared policy distribution, referred to as the \textit{world policy}, which enables rapid adaptation to new tasks while retaining valuable knowledge from previous experiences. Our theoretical analysis establishes a relationship between the algorithm's generalization performance and the number of prior tasks preserved in memory. We also derive the sample complexity of EPIC in terms of RL regret. Extensive experiments on a variety of environments demonstrate that EPIC significantly outperforms existing methods in lifelong RL, offering both theoretical guarantees and practical efficacy through the use of the world policy.

Large language models (LLMs) have shown remarkable versatility across tasks, but aligning them with individual human preferences remains challenging due to the complexity and diversity of these preferences. Existing methods often overlook the fact that preferences are multi-objective, diverse, and hard to articulate, making full alignment difficult. In response, we propose an active preference learning framework that uses binary feedback to estimate user preferences across multiple objectives. Our approach leverages Bayesian inference to update preferences efficiently and reduces user feedback through an acquisition function that optimally selects queries. Additionally, we introduce a parameter to handle feedback noise and improve robustness. We validate our approach through theoretical analysis and experiments on language generation tasks, demonstrating its feedback efficiency and effectiveness in personalizing model responses.

In this work, we explore modeling change points in time-series data using neural stochastic differential equations (neural SDEs). We propose a novel model formulation and training procedure based on the variational autoencoder (VAE) framework for modeling time-series as a neural SDE. Unlike existing algorithms training neural SDEs as VAEs, our proposed algorithm only necessitates a Gaussian prior of the initial state of the latent stochastic process, rather than a Wiener process prior on the entire latent stochastic process. We develop two methodologies for modeling and estimating change points in time-series data with distribution shifts. Our iterative algorithm alternates between updating neural SDE parameters and updating the change points based on either a maximum likelihood-based approach or a change point detection algorithm using the sequential likelihood ratio test. We provide a theoretical analysis of this proposed change point detection scheme. Finally, we present an empirical evaluation that demonstrates the expressive power of our proposed model, showing that it can effectively model both classical parametric SDEs and some real datasets with distribution shifts.