arXiv AI
Jul 23
Sophisticated Policies from Epistemic Priors
arXiv:2607. 19518v1 Announce Type: new Abstract: Sophisticated Inference is a variant of active inference often associated with recursive belief modeling and tree search.
Parameter Learning in Active Inference via Variational Message Passing
The paper extends active inference to incorporate learning of observation and transition parameters by assigning each factor its own Dirichlet prior, thereby preserving information‑gain corrections that vanish under standard mean‑field factorization. This approach yields a closed‑form objective whose stationary point is solved by sum‑product message passing with augmented factor functions. Experiments on two grid POMDPs demonstrate that the resulting planner reliably learns the model and reaches the goal in every episode, outperforming planners that lack this learning capability.
Related stories
arXiv AI
Jun 4
What Type of Inference is Active Inference?
arXiv:2606. 04935v1 Announce Type: new Abstract: Active inference casts decision-making as inference, with the Expected Free Energy (EFE) unifying goal-directed and information-seeking behavior.
arXiv AI
Jul 21
Uncovering Latent Reasoning Strategies in Language Models
arXiv:2607. 17674v1 Announce Type: cross Abstract: A language model $p_\theta(y \mid x)$ trained on reasoning tasks learns to solve problems via multiple distinct strategies, yet these strategies are implicit and entangled within the model's response distribution.
arXiv AI
Oct 15
Belief Embedding Tree Search
arXiv:2610.15046v1 Announce Type: new Abstract: Beliefs in a Partially Observable Markov Decision Process (POMDP) are posterior distributions over the hidden state given the agent's history of action...
arXiv AI
Aug 19
Expected free energy as an information constraint on the Bethe Lagrangian
The paper introduces a Bethe Lagrangian formulation of expected free energy (EFE) that preserves a Kullback–Leibler structure, enabling message‑passing inference. By imposing an information constraint—requiring the mutual information between future observations, states, and parameters given actions to be at least the entropy of the goal prior—the authors recover the standard EFE solution at a specific Karush‑Kuhn‑Tucker multiplier. They analyze how varying this multiplier transitions the agent’s epistemic drive through inactive, interior, and saturated regimes, and benchmark the constrained Bethe agent against EFE and Q‑MDP on three tasks.
arXiv Machine Learning
Jul 7
Direct Bethe Free Energy Minimization for Bayesian Neural Networks
arXiv:2605. 08446v3 Announce Type: replace Abstract: Bayesian neural networks are typically trained against the evidence lower bound (ELBO), whose Jensen gap closes only when the variational posterior is exact.
arXiv Statistics ML
Sep 24
BOCO: Bayesian Online Contextual Optimization for Decision-Focused Online Learning
arXiv:2511.20413v2 Announce Type: replace-cross Abstract: \emph{Decision-focused learning} (DFL) trains predictive models to optimize downstream decisions rather than prediction accuracy alone. While...
arXiv Machine Learning
Jun 9
Reformulate LLM Reinforcement Learning for Efficient Training under Black-box Discrepancy
arXiv:2606. 08779v1 Announce Type: new Abstract: Reinforcement Learning (RL) has emerged as a pivotal post-training paradigm, yet it frequently suffers from unpredictable sub-optimum performance or even training collapses.
arXiv Machine Learning
Jul 7
Gradient Descent as Implicit EM in Distance-Based Neural Models
arXiv:2512. 24780v2 Announce Type: replace Abstract: Neural networks trained with standard objectives exhibit behaviors characteristic of probabilistic inference: soft clustering, prototype specialization, and Bayesian uncertainty tracking.
arXiv Machine Learning
Aug 19
The concentration game: Bayesian updating, regret, and information
The paper introduces a two-player zero-sum repeated game between a learner and nature that simultaneously captures Bayesian updating and an exact decomposition of exponential-weights regret. The game’s terminal payoff reflects the maximum gain a comparator can achieve given a fixed relative entropy from the prior, while the one-step constraint limits nature’s move by an information budget. The resulting regret splits into three precise components—per-round information loss, an additive retempering drift, and the comparator’s information relative to the prior—providing a unified framework that explains concentration phenomena, large-deviation bounds, and various learning methods such as bandits, posterior sampling, aggregation, and boosting.
arXiv AI
Jul 21
Expected Free Energy as Belief-Dependent Utility for rho-POMDPs
arXiv:2607. 16981v1 Announce Type: new Abstract: An agent acting under partial observability must decide when to gather information and which observations are worth their cost.
arXiv AI
Jul 16
The SIGReg Objective as Variational Free Energy: A Theoretical Active-Inference Account of JEPA World Models
arXiv:2607. 13612v1 Announce Type: cross Abstract: Joint-Embedding Predictive Architectures (JEPAs) are the dominant design for latent world models, yet they are usually justified by empirical performance rather than a normative principle.