PAWA tackles multi-agent world modeling in shared environments like Minecraft, where one player's actions change what every other player sees—not just in their local field of view, but in directions they are not currently looking. The core idea is to jointly forecast each player's point-of-view (POV) video alongside a world-stabilized panoramic video, conditioned on all players' actions. The model uses a dual-tower architecture with ray-aligned cross-attention that links tokens viewing the same world direction regardless of whether they appear in rectilinear POV or equirectangular panorama coordinates, plus a quadratic-phase player encoding that scales identities as $m^2$ instead of $m+1$. The work also introduces a scalable dataset pipeline that records gameplay without graphics clients and renders replays offline, supporting more than 64 simultaneous players.

Momentum is a pillar of modern optimization, yet its benefits in reinforcement learning with verifiable rewards (RLVR) remain unclear. This paper studies a KL-regularized bandit formulation of RLVR and proves that, with exact policy gradients, larger EMA coefficients strictly accelerate convergence within a safe overshoot-free range. Its central insight is that stochastic sampling obscures this gain unless the coefficient scales with the learning rate so that $\eta/(1-\beta)$ stays constant, a principle validated in bandits and in LLM post-training on Qwen models.

Post-legalization placement refinement seeks to reduce wirelength after global placement without breaking site constraints. This paper audits whether a GNN-PPO agent can handle that task by systematically exploring 18 training configurations on three ISPD2015 benchmarks, varying rewards, action spaces, GNN architectures, and graph-partitioning strategies. The study finds that RL lags behind classical tools, and it dissects why—pinning blame on credit-assignment bottlenecks, GNN over-smoothing, and netlist regularity—while proposing a lightweight netlist classifier to guide training decisions.

This paper asks whether releasing learned embeddings from pretrained brain MRI encoders—instead of raw scans—actually protects patient privacy. The authors audit five diverse encoders on two longitudinal datasets, measuring patient linkage, re-identification, and sex-attribute inference under an explicit attacker model. They find that several embeddings retain enough subject-specific structure to enable rank-1 re-identification above 90% on both datasets, and that sex can be inferred with balanced accuracy up to 95.5%. The core takeaway is that embedding release alone is insufficient to establish anonymity.

This paper reframes backdoor stealthiness in deep neural networks as a multi-space measurement problem rather than a single latent property. The authors benchmark ten attacks across four datasets and three architectures, measuring observable footprints in input, feature, and parameter space to show that stealthiness is fragmented: an attack can be quiet in one space and conspicuous in another. They also introduce two diagnostic metrics—CDBI and TUP—to address blind spots in global feature clustering and channel-level concentration. The central message is that future stealthiness claims should be backed by evidence from all three spaces, not by a single metric or defense.

Semi-gradient TD with nonlinear function approximation is notoriously unstable because the bootstrap target moves as the parameters evolve. This paper studies the problem in the mean-field regime for two-layer neural networks and identifies a stability regime where entropy-driven relaxation dominates the sensitivity of the moving target. Under the resulting condition, the authors prove exponential last-iterate convergence and quantify the regularization bias, then extend the same mechanism to soft Q-learning.

MeanFlow normally couples the prediction horizon $h_k$ to the update step $\delta_k$. This paper decouples them, showing on CIFAR-10 that longer horizons hurt FID near the source $(t=0)$ but help at the midpoint $(t=0.5)$, and that this midpoint gain depends on earlier guidance choices. A frozen four-query rule improves mean FID by 2.0 without retraining or extra calls, yet ImageNet calibration finds no benefit beyond shifting query starts. The work establishes prediction horizon as a distinct but configuration-sensitive inference control.

Multi-party floor control is the next bottleneck for full-duplex voice agents: systems must decide not just when speech ends, but whether overlapping listeners are backchanneling or genuinely taking the floor. MultiPartyTurnBench introduces 240 synthetically generated, channel-separated conversations spanning two to five speakers and six interaction types, evaluating End-of-Turn (EOT) and Interruption (INT) detection with exact causal annotations. The results reveal a stark capability asymmetry: turn completion is largely tractable, while selective interruption handling remains near chance even for state-of-the-art commercial agents.

Agentic AI pipelines interleave LLM decisions, state changes, tool calls, and cross-agent delegation, so a single monitoring formalism cannot cover every boundary. This paper proposes AgenticRV, an architecture that separates execution-boundary adapters from monitors via canonical events, per-monitor projections and abstractions, and a policy layer that arbitrates normalized verdicts. Its core contribution is an interoperability contract with formal conditions for when observations suffice, when monitors can be replaced without changing decisions, and when blocking actually prevents effects.

CPS security analysts operate under need-to-know constraints that give them incomplete cross-domain views, risking flawed risk assessments. AWARE-RG addresses this by building a role-bounded abstraction lattice over Resiliency Graphs using MITRE ATT&CK tactics ordered by the ICS Cyber Kill Chain. It uses AI planning to validate attack-fault queries, repair incomplete analyst paths, and generate need-to-know explanations without exposing unauthorized predicates.

ReproBAIT evaluates whether sandboxed coding agents can reconstruct published biological AI tools from limited text descriptions when supplied with training data and a fixed compute budget. The authors task frontier LLMs with replicating three tools—a closed-source molecular-design model, ProteinMPNN, and RFdiffusion3—and score functional parity on held-out data rather than source-code similarity. Near-parity is reached on the two simpler tools for only a few dollars, while the most complex tool defeats all agents within the allotted budget, raising questions about memorisation and whether withholding code is enough to control access.

Theory-of-Mind reasoning requires connecting an agent's perception to their belief and then to their action, yet large language models often fail to compose these inferences even when each part is known. This paper proposes rule-based reinforcement learning: using Group Relative Policy Optimization to train models to emit concise if–then rules that bridge perception, belief, and action before answering. The authors show that this approach outperforms free-form chain-of-thought reasoning and supervised fine-tuning across four ToM benchmarks, while using roughly half the tokens, and that rules from larger models improve downstream GPT-4o-mini accuracy.

Reinforcement learning for language agents suffers from sparse trajectory-level rewards that do not directly supervise environment dynamics. DGS-WM addresses this by introducing Decision-Guided Selective World Modeling: instead of forcing the policy to predict entire environment observations as auxiliary targets, it selects only observation units that support the subsequent policy response and discounts already-mastered targets via confidence-adaptive weighting. This selective co-training yields consistent gains over full-observation counterparts across QA and web-shopping benchmarks.

StateCraft tackles the long-horizon inconsistency problem in video world models by explicitly decoupling a persistent, structured abstract state from the visual rendering pipeline. A finetuned vision-language model maintains and updates a graph-structured abstract state in response to actions, including off-screen changes, while a video diffusion model generates each frame conditioned on that state. This enables consistent revisits, direct state editing, and large empirical gains on a new GUI-world-model benchmark measuring memory and controllability.

The paper asks how much meaning language models can actually share through language alone. To measure this, the authors introduce two tools — an epsilon-estimation algorithm that finds near-optimal prompts and estimates how much performance is left on the table, and a reconstruction bottleneck that scores how well a receiver model recovers a sender model's concept from labeled examples alone. They then apply observational scaling laws across 1.2k qualitative concepts in eight domains, classifying each as already articulable ('reaches'), potentially articulable with stronger models ('rising'), or permanently tacit ('bounded').

Decentralised HVAC control in multi-zone buildings is difficult because each zone controller sees only local state while the zones share a single air loop. This paper assigns every zone an independent active-inference agent that learns its own transition model online, infers hidden coupling via a latent disturbance factor, and adjusts its energy weight in response to a live congestion signal drawn from the metered power of other floors, with no inter-agent messages. On five noisy test years the fleet achieves the lowest mean building peak ($88.3~\mathrm{kW}$) and the best coincidence factor ($0.929$) among nine controllers, while running the full fleet on a Raspberry Pi 5 with a p95 control latency of $442~\mathrm{ms}$.

Most rule-reasoning benchmarks present scenarios as a few flat facts, making it easy to spot relevant rules. RuleTrek instead distributes scenario evidence across heterogeneous files and decomposes agentic reasoning into Rule Identification (RI), Rule Deduction (RD), and End-to-End (E2E) evaluation. Experiments on 11 frontier models show that even the best agent achieves only 62.47% RI and 60.43% RD, while E2E caps at 55.29%, suggesting that distributed, multi-file scenarios pose severe challenges for current agents.

L3M asks how to run AI agents so that useful computation scales safely across days, machines, and model calls. It answers with a Git-backed, distributed runtime written in Lean 4 where capability tokens confine every tool call, trajectories are durable append-only logs, and agents self-extend by writing sealed programs whose types bound their authority. The goal is a harness where trust narrows rather than widens as agents grow more capable.

LiFi targets the task of generating metric LiDAR point clouds from cheap onboard multi-view camera images for autonomous driving. The core idea is a latent diffusion model guided by two decoupled branches: a geometry branch that aligns dense monocular depth estimates to the target LiDAR scanning pattern via a reverse sampling algorithm and uncertainty-modulated encoding, and a semantic branch that aggregates global cross-view context. If the approach holds up, it could enable large-scale data augmentation, cross-modal simulation, and sensor completion without requiring expensive LiDAR capture at deployment time.

Multi-turn language agents trained with trajectory-level rewards frequently suffer from misguided reinforcement: tokens inside undesirable behaviors can still receive positive advantages. The paper proposes Confidence-Guided Credit Correction (C3), a plug-in operator that selectively reverses positive advantages within detector-localized undesirable spans, scaling the correction by the behavior policy's confidence. Because C3 leaves rewards, returns, and advantage estimators untouched, it is compatible with both critic-based and critic-free optimizers, offering a lightweight and general form of process supervision.