A Collision-Free Hot-Tier Extension for Engram-Style Conditional Memory: A Controlled Study of Training Dynamics

We investigate whether high-frequency key collisions are a primary bottleneck in Engram-style conditional memory. To isolate the effect of collisions, we introduce Engram-Nine, a collision-free hot-tier extension that maps the most frequent n-grams through a Minimal Perfect Hash Function (MPHF) while retaining the original multi-head hashed lookup as a cold tier. Under a strictly iso-parameter setup, the collision-free design does not consistently improve validation loss. Through route-stratified evaluation (decomposing per-token loss into hot/cold contributions), we uncover a consistent "hot-to-cold advantage flip" during training: hot (high-frequency) positions initially have lower loss, but cold positions eventually surpass them. Crucially, collision-free configurations flip earlier than collision-prone baselines, suggesting that collisions act as implicit regularization. We also identify a gating mismatch: the gate learns to favor hot positions early in training, but this preference persists even after the flip, assigning higher weights to positions with higher loss. Our findings suggest that improving lookup precision alone does not guarantee better training outcomes. The dominant limitation may lie in gating credit assignment rather than index accuracy, and collision-induced noise may provide beneficial regularization that should not be naively eliminated.

Optimizing Decoding Paths in Masked Diffusion Models by Quantifying Uncertainty

Masked Diffusion Models (MDMs) offer flexible, non-autoregressive generation, but this freedom introduces a challenge: final output quality is highly sensitive to the decoding order. We are the first to formalize this issue, attributing the variability in output quality to the cumulative predictive uncertainty along a generative path. To quantify this uncertainty, we introduce Denoising Entropy, a computable metric that serves as an internal signal for evaluating generative process. Leveraging this metric, we propose two algorithms designed to optimize the decoding path: a post-hoc selection method and a real-time guidance strategy. Experiments demonstrate that our entropy-guided methods significantly improve generation quality, consistently boosting accuracy on challenging reasoning, planning, and code benchmarks. Our work establishes Denoising Entropy as a principled tool for understanding and controlling generation, effectively turning the uncertainty in MDMs from a liability into a key advantage for discovering high-quality solutions.

Rethinking Expert Trajectory Utilization in LLM Post-training

While effective post-training integrates Supervised Fine-Tuning (SFT) and Reinforcement Learning (RL), the optimal mechanism for utilizing expert trajectories remains unresolved. We propose the Plasticity-Ceiling Framework to theoretically ground this landscape, decomposing performance into foundational SFT performance and the subsequent RL plasticity. Through extensive benchmarking, we establish the Sequential SFT-then-RL pipeline as the superior standard, overcoming the stability deficits of synchronized approaches. Furthermore, we derive precise scaling guidelines: (1) Transitioning to RL at the SFT Stable or Mild Overfitting Sub-phase maximizes the final ceiling by securing foundational SFT performance without compromising RL plasticity; (2) Refuting ``Less is More'' in the context of SFT-then-RL scaling, we demonstrate that Data Scale determines the primary post-training potential, while Trajectory Difficulty acts as a performance multiplier; and (3) Identifying that the Minimum SFT Validation Loss serves as a robust indicator for selecting the expert trajectories that maximize the final performance ceiling. Our findings provide actionable guidelines for maximizing the value extracted from expert trajectories.

Evo-1: Lightweight Vision-Language-Action Model with Preserved Semantic Alignment

Vision-Language-Action (VLA) models have emerged as a powerful framework that unifies perception, language, and control, enabling robots to perform diverse tasks through multimodal understanding. However, current VLA models typically contain massive parameters and rely heavily on large-scale robot data pretraining, leading to high computational costs during training, as well as limited deployability for real-time inference. Moreover, most training paradigms often degrade the perceptual representations of the vision-language backbone, resulting in overfitting and poor generalization to downstream tasks. In this work, we present Evo-1, a lightweight VLA model that reduces computation and improves deployment efficiency, while maintaining strong performance without pretraining on robot data. Evo-1 builds on a native multimodal Vision-Language model (VLM), incorporating a novel cross-modulated diffusion transformer along with an optimized integration module, together forming an effective architecture. We further introduce a two-stage training paradigm that progressively aligns action with perception, preserving the representations of the VLM. Notably, with only 0.77 billion parameters, Evo-1 achieves state-of-the-art results on the Meta-World and RoboTwin suite, surpassing the previous best models by 12.4% and 6.9%, respectively, and also attains a competitive result of 94.8% on LIBERO. In real-world evaluations, Evo-1 attains a 78% success rate with high inference frequency and low memory overhead, outperforming all baseline methods. We release code, data, and model weights to facilitate future research on lightweight and efficient VLA models.

Stop Wasting Your Tokens: Towards Efficient Runtime Multi-Agent Systems

While Multi-Agent Systems (MAS) excel at complex tasks, their growing autonomy with operational complexity often leads to critical inefficiencies, such as excessive token consumption and failures arising from misinformation. Existing methods primarily focus on post-hoc failure attribution, lacking proactive, real-time interventions to enhance robustness and efficiency. To this end, we introduce SupervisorAgent, a lightweight and modular framework for runtime, adaptive supervision that operates without altering the base agent's architecture. Triggered by an LLM-free adaptive filter, SupervisorAgent intervenes at critical junctures to proactively correct errors, guide inefficient behaviors, and purify observations. On the challenging GAIA benchmark, SupervisorAgent reduces the token consumption of the Smolagent framework by an average of 29.45% without compromising its success rate. Extensive experiments across five additional benchmarks (math reasoning, code generation, and question answering) and various SoTA foundation models validate the broad applicability and robustness of our approach. The code is available at https://github.com/LINs-lab/SupervisorAgent.

Learning to Play Multi-Follower Bayesian Stackelberg Games

In a multi-follower Bayesian Stackelberg game, a leader plays a mixed strategy over $L$ actions to which $n\ge 1$ followers, each having one of $K$ possible private types, best respond. The leader's optimal strategy depends on the distribution of the followers' private types. We study an online learning version of this problem: a leader interacts for $T$ rounds with $n$ followers with types sampled from an unknown distribution every round. The leader's goal is to minimize regret, defined as the difference between the cumulative utility of the optimal strategy and that of the actually chosen strategies. We design learning algorithms for the leader under different feedback settings. Under type feedback, where the leader observes the followers' types after each round, we design algorithms that achieve $\mathcal O\big(\sqrt{\min\{L\log(nKA T), nK \} \cdot T} \big)$ regret for independent type distributions and $\mathcal O\big(\sqrt{\min\{L\log(nKA T), K^n \} \cdot T} \big)$ regret for general type distributions. Interestingly, those bounds do not grow with $n$ at a polynomial rate. Under action feedback, where the leader only observes the followers' actions, we design algorithms with $\mathcal O( \min\{\sqrt{ n^L K^L A^{2L} L T \log T}, K^n\sqrt{ T } \log T \} )$ regret. We also provide a lower bound of $\Omega(\sqrt{\min\{L, nK\}T})$, almost matching the type-feedback upper bounds.

WOMAC: A Mechanism For Prediction Competitions

Competitions are widely used to identify top performers in judgmental forecasting and machine learning, and the standard competition design ranks competitors based on their cumulative scores against a set of realized outcomes or held-out labels. However, this standard design is neither incentive-compatible nor very statistically efficient. The main culprit is noise in outcomes/labels that experts are scored against; it allows weaker competitors to often win by chance, and the winner-take-all nature incentivizes misreporting that improves win probability even if it decreases expected score. Attempts to achieve incentive-compatibility rely on randomized mechanisms that add even more noise in winner selection, but come at the cost of determinism and practical adoption. To tackle these issues, we introduce a novel deterministic mechanism: WOMAC (Wisdom of the Most Accurate Crowd). Instead of scoring experts against noisy outcomes, as is standard, WOMAC scores experts against the best ex-post aggregate of peer experts' predictions given the noisy outcomes. WOMAC is also more efficient than the standard competition design in typical settings. While the increased complexity of WOMAC makes it challenging to analyze incentives directly, we provide a clear theoretical foundation to justify the mechanism. We also provide an efficient vectorized implementation and demonstrate empirically on real-world forecasting datasets that WOMAC is a more reliable predictor of experts' out-of-sample performance relative to the standard mechanism. WOMAC is useful in any competition where there is substantial noise in the outcomes/labels.

Learning to Coordinate Bidders in Non-Truthful Auctions

In non-truthful auctions such as first-price and all-pay auctions, the independent strategic behaviors of bidders, with the corresponding equilibrium notion -- Bayes Nash equilibria -- are notoriously difficult to characterize and can cause undesirable outcomes. An alternative approach to designing better auction systems is to coordinate the bidders: let a mediator make incentive-compatible recommendations of correlated bidding strategies to the bidders, namely, implementing a Bayes correlated equilibrium (BCE). The implementation of BCE, however, requires knowledge of the distribution of bidders' private valuations, which is often unavailable. We initiate the study of the sample complexity of learning Bayes correlated equilibria in non-truthful auctions. We prove that the BCEs in a large class of non-truthful auctions, including first-price and all-pay auctions, can be learned with a polynomial number $\tilde O(\frac{n}{\varepsilon^2})$ of samples from the bidders' value distributions. Our technique is a reduction to the problem of estimating bidders' expected utility from samples, combined with an analysis of the pseudo-dimension of the class of all monotone bidding strategies of bidders.

CPathAgent: An Agent-based Foundation Model for Interpretable High-Resolution Pathology Image Analysis Mimicking Pathologists' Diagnostic Logic

Recent advances in computational pathology have led to the emergence of numerous foundation models. However, these approaches fail to replicate the diagnostic process of pathologists, as they either simply rely on general-purpose encoders with multi-instance learning for classification or directly apply multimodal models to generate reports from images. A significant limitation is their inability to emulate the diagnostic logic employed by pathologists, who systematically examine slides at low magnification for overview before progressively zooming in on suspicious regions to formulate comprehensive diagnoses. To address this gap, we introduce CPathAgent, an innovative agent-based model that mimics pathologists' reasoning processes by autonomously executing zoom-in/out and navigation operations across pathology images based on observed visual features. To achieve this, we develop a multi-stage training strategy unifying patch-level, region-level, and whole-slide capabilities within a single model, which is essential for mimicking pathologists, who require understanding and reasoning capabilities across all three scales. This approach generates substantially more detailed and interpretable diagnostic reports compared to existing methods, particularly for huge region understanding. Additionally, we construct an expert-validated PathMMU-HR$^{2}$, the first benchmark for huge region analysis, a critical intermediate scale between patches and whole slides, as diagnosticians typically examine several key regions rather than entire slides at once. Extensive experiments demonstrate that CPathAgent consistently outperforms existing approaches across three scales of benchmarks, validating the effectiveness of our agent-based diagnostic approach and highlighting a promising direction for the future development of computational pathology.

PiFlow: Principle-aware Scientific Discovery with Multi-Agent Collaboration

Large Language Model (LLM)-based multi-agent systems (MAS) demonstrate remarkable potential for scientific discovery. Existing approaches, however, often automate scientific discovery using predefined workflows that lack rationality constraints. This often leads to aimless hypothesizing and a failure to consistently link hypotheses with evidence, thereby hindering systematic uncertainty reduction. Overcoming these limitations fundamentally requires systematic uncertainty reduction. We introduce \texttt{PiFlow}, an information-theoretical framework, treating automated scientific discovery as a structured uncertainty reduction problem guided by principles (e.g., scientific laws). In evaluations across three distinct scientific domains -- discovering nanomaterial structures, bio-molecules, and superconductor candidates with targeted properties -- our method significantly improves discovery efficiency, reflected by a 73.55\% increase in the Area Under the Curve (AUC) of property values versus exploration steps, and enhances solution quality by 94.06\% compared to a vanilla agent system. Overall, \texttt{PiFlow} serves as a Plug-and-Play method, establishing a novel paradigm shift in highly efficient automated scientific discovery, paving the way for more robust and accelerated AI-driven research. Code is publicly available at our \href{https://github.com/amair-lab/PiFlow}{GitHub}.