5arXiv cs.LG (Machine Learning)·
Latent Memory Palace: Variational inference framework enables adaptive reasoning for continuous control policies
Researchers introduce Latent Memory Palace (LMP), a method that formulates reasoning for continuous control policies as variational inference over an autoregressive latent distribution, analogous to a memory palace. The approach derives a latent-space reinforcement learning technique to optimize the variational lower bound, yielding a policy (LMP-π) with adaptive test-time compute allocation and a variable-length action tokenizer (LMP-tok) that improves downstream autoregressive policies. The work addresses the gap between language model chain-of-thought reasoning and continuous control, where language-space reasoning lacks spatial granularity. Results are demonstrated in both simulation and real-world domains.
Related guides (2)
Related events (8)
Reasoning in Memory (RiM): Latent Reasoning via Working Memory Blocks in LLMs
RiM introduces a latent reasoning method that replaces autoregressive chain-of-thought token generation with fixed sequences of special 'memory block' tokens, allowing LLMs to perform internal computation without externalizing intermediate steps. These memory blocks are processed in a single forward pass rather than generated autoregressively, improving compute efficiency at test time. Training uses a two-stage curriculum: first grounding memory blocks by predicting explicit reasoning steps, then discarding step-level supervision and refining answers iteratively. Experiments across multiple model families and sizes show RiM matches or exceeds existing latent reasoning methods.
Mem-π: Adaptive Memory for LLM Agents via On-Demand Generation and Decoupled RL
Mem-π introduces a framework where a dedicated language or vision-language model generates context-specific guidance for LLM agents on demand, rather than retrieving static entries from episodic memory banks. The system is trained with a decision-content decoupled reinforcement learning objective that jointly learns when to generate guidance and what to generate, enabling abstention when generation would not help. Evaluated across web navigation, terminal-based tool use, and text-based embodied interaction benchmarks, Mem-π achieves over 30% relative improvement on web navigation tasks compared to retrieval-based and prior RL-optimized memory baselines.
RECALL: Active continual learning for Vision-Language-Action models via uncertainty-guided recovery data collection
Researchers propose RECALL, an active continual learning paradigm for Vision-Language-Action (VLA) robot models that uses uncertainty-guided data collection to target states where the policy struggles, rather than passively collecting demonstrations after failures. The paper demonstrates improved fine-tuning efficiency over passive imitation learning but identifies catastrophic forgetting as a key challenge when incorporating recovery data. The authors evaluate continual learning mitigations including replay-based data mixing and elastic weight consolidation, characterizing tradeoffs between plasticity and retention in large autoregressive robot policies.
AutoMem: Automated framework trains LLMs to manage memory as a learnable cognitive skill
AutoMem is a new framework that treats memory management in LLMs as a trainable skill, using two optimization loops: one that iteratively revises memory structure via trajectory review by a strong LLM, and one that distills good memory decisions into direct training signal for the agent model. Evaluated on three long-horizon procedurally generated games (Crafter, MiniHack, NetHack), optimizing memory alone yielded 2x-4x performance improvements, bringing a 32B open-weight model competitive with frontier systems like Claude Opus 4.5 and Gemini 3.1 Pro Thinking. The work draws on cognitive science concepts of metamemory and demonstrates that memory management is an independently learnable, high-leverage capability for long-horizon agentic tasks.
AIR: Adaptive Interleaved Reasoning with Code in Multimodal LLMs via Reinforcement Learning
Researchers propose AIR, a system that trains multimodal large language models to adaptively interleave reasoning with code execution for numerical computation tasks, going beyond prior work that focused only on visual operations. The approach combines a two-stage cold-start data pipeline, RL dataset filtering, and a group-constrained reward function for tool-invocation decisions. Experiments show a 6.1 percentage point average improvement on evaluation benchmarks, with interleaved reasoning samples gaining 9.9 pp and tool-use success exceeding 95%.
ACTS: Agentic Chain-of-Thought Steering for efficient and controllable LLM reasoning
Researchers introduce Agentic Chain-of-Thought Steering (ACTS), a framework that formulates inference-time reasoning control as a Markov decision process, where a controller agent adaptively steers a frozen reasoner by issuing reasoning strategy directives and steering phrases at each step. The controller is initialized from synthetic steering trajectories with multi-budget augmentation and further optimized via reinforcement learning with budget-conditioned reward shaping. ACTS matches full-thinking performance with significant token savings and enables controllable accuracy-efficiency trade-offs across multiple benchmarks and reasoner models.
ACPO: Adaptive Clip Policy Optimization improves RLVR training for LLM reasoning
A new arXiv preprint provides theoretical analysis of Reinforcement Learning from Verifiable Rewards (RLVR) updates, identifying off-policy degree and gradient expectation as key factors governing update dynamics. The authors show that differences in gradient steps per rollout substantially affect importance sampling ratio distributions and which tokens dominate updates. Based on this analysis, they propose Adaptive Clip Policy Optimization (ACPO), which adjusts clipping boundaries per token group by empirical variance of importance sampling ratios, outperforming DAPO and CISPO baselines on 3B and 7B models across math, tabular QA, and logic benchmarks.
IA-VQC-DPC: Intervention-aware quantum predictive control with safety attribution for learned policies
A new arXiv preprint introduces Intervention-Aware Variational Quantum Differentiable Predictive Control (IA-VQC-DPC), a framework that trains variational quantum circuit policies under a primal-dual intervention budget to penalize over-reliance on downstream safety filters (Control-Barrier-Function projections). The work also proposes a safety-attribution protocol that decomposes trajectory corrections into policy-level versus filter-level contributions, enabling measurement of whether a policy has genuinely learned safe behavior or is merely being silently repaired by its safety layer. Experiments on BOPTEST building-control emulators show the quantum policy achieves significantly lower pre-filter violations than a matched classical policy at equal parameter budget, with a notable negative result: a learned energy head is only safe when paired with a distribution-aware runtime guard.