SOP: A Scalable Online Post-Training System for Vision-Language-Action Models

Vision-language-action (VLA) models achieve strong generalization through large-scale pre-training, but real-world deployment requires expert-level task proficiency in addition to broad generality. Existing post-training approaches for VLA models are typically offline, single-robot, or task-specific, limiting effective on-policy adaptation and scalable learning from real-world interaction. We introduce a Scalable Online Post-training (SOP) system that enables online, distributed, multi-task post-training of generalist VLA models directly in the physical world. SOP tightly couples execution and learning through a closed-loop architecture in which a fleet of robots continuously streams on-policy experience and human intervention signals to a centralized cloud learner, and asynchronously receives updated policies. This design supports prompt on-policy correction, scales experience collection through parallel deployment, and preserves generality during adaptation. SOP is agnostic to the choice of post-training algorithm; we instantiate it with both interactive imitation learning (HG-DAgger) and reinforcement learning (RECAP). Across a range of real-world manipulation tasks including cloth folding, box assembly, and grocery restocking, we show that SOP substantially improves the performance of large pretrained VLA models while maintaining a single shared policy across tasks. Effective post-training can be achieved within hours of real-world interaction, and performance scales near-linearly with the number of robots in the fleet. These results suggest that tightly coupling online learning with fleet-scale deployment is instrumental to enabling efficient, reliable, and scalable post-training of generalist robot policies in the physical world.

Movable Antenna Enhanced Multi-Region Beam Coverage: A Multi-Notch-Filter-Inspired Design

Movable antenna (MA) has emerged as a promising technology to enhance wireless communication performance by exploiting the new degree of freedom (DoF) via antenna position optimization. In this letter, we investigate the MA-enhanced wide beam coverage over multiple subregions in the spatial domain. Specifically, we aim to maximize the minimum beam gain over the desired subregions by jointly optimizing the transmit beamforming and antenna position vector (APV). Although this problem is non-convex, we propose an efficient algorithm to solve it by leveraging the similarity between the considered multi-region coverage and classical multi-notch filter (MNF) design. In particular, we construct a spatial MNF-based transmit beamforming vector by assuming a continuous amplitude and phase-shift profile within the antenna movement region. Based on this continuous profile, we propose a sequential update algorithm to select an optimal subset of MA positions for multi-region coverage, jointly with a Gibbs sampling (GS) procedure to avoid undesired local optimum. Numerical results show that our proposed algorithm can significantly outperform conventional fixed position antennas (FPAs) and achieve a comparable performance to the alternating optimization (AO) algorithm with dramatically lower complexity.

Large Language Newsvendor: Decision Biases and Cognitive Mechanisms

Problem definition: Although large language models (LLMs) are increasingly integrated into business decision making, their potential to replicate and even amplify human cognitive biases cautions a significant, yet not well-understood, risk. This is particularly critical in high-stakes operational contexts like supply chain management. To address this, we investigate the decision-making patterns of leading LLMs using the canonical newsvendor problem in a dynamic setting, aiming to identify the nature and origins of their cognitive biases. Methodology/results: Through dynamic, multi-round experiments with GPT-4, GPT-4o, and LLaMA-8B, we tested for five established decision biases. We found that LLMs consistently replicated the classic ``Too Low/Too High'' ordering bias and significantly amplified other tendencies like demand-chasing behavior compared to human benchmarks. Our analysis uncovered a ``paradox of intelligence'': the more sophisticated GPT-4 demonstrated the greatest irrationality through overthinking, while the efficiency-optimized GPT-4o performed near-optimally. Because these biases persist even when optimal formulas are provided, we conclude they stem from architectural constraints rather than knowledge gaps. Managerial implications: First, managers should select models based on the specific task, as our results show that efficiency-optimized models can outperform more complex ones on certain optimization problems. Second, the significant amplification of bias by LLMs highlights the urgent need for robust human-in-the-loop oversight in high-stakes decisions to prevent costly errors. Third, our findings suggest that designing structured, rule-based prompts is a practical and effective strategy for managers to constrain models' heuristic tendencies and improve the reliability of AI-assisted decisions.

Rotatable Antenna Array-Enhanced Null Steering: Performance Analysis and Optimization

Conventional fixed-orientation antenna (FOA) arrays offer limited degrees of freedom (DoF) for flexible beamforming such as null steering. To address this limitation, we propose a new rotatable antenna array (RAA) architecture in this paper, which enables three-dimensional (3D) rotational control of an antenna array to provide enhanced spatial flexibility for null steering. To characterize its performance, we aim to jointly optimize the 3D rotational angles of the RAA, to maximize the beam gain over a given desired direction, while nulling those over multiple interference directions under zero-forcing (ZF) beamforming. However, this problem is non-convex and challenging to tackle due to the highly nonlinear expression of the beam gain in terms of the rotational angles. To gain insights, we first examine several special cases including both isotropic and directional antenna radiation patterns, deriving the conditions under which full beam gain can be achieved over the desired direction while meeting the nulling constraints for interference directions. These conditions clearly indicate that compared with FOA arrays, RAAs can significantly relax the angular separation requirement for achieving effective null steering. For other general cases, we propose a sequential update algorithm, that iteratively refines the 3D rotational angles by discretizing the 3D angular search space. To avoid undesired local optimum, a Gibbs sampling (GS) procedure is also employed between two consecutive rounds of sequential update for solution exploration. Simulation results verify our analytical results and show superior null-steering performance of RAAs to FOA arrays.

Fine-Grained Zero-Shot Learning with Attribute-Centric Representations

Recognizing unseen fine-grained categories demands a model that can distinguish subtle visual differences. This is typically achieved by transferring visual-attribute relationships from seen classes to unseen classes. The core challenge is attribute entanglement, where conventional models collapse distinct attributes like color, shape, and texture into a single visual embedding. This causes interference that masks these critical distinctions. The post-hoc solutions of previous work are insufficient, as they operate on representations that are already mixed. We propose a zero-shot learning framework that learns AttributeCentric Representations (ACR) to tackle this problem by imposing attribute disentanglement during representation learning. ACR is achieved with two mixture-of-experts components, including Mixture of Patch Experts (MoPE) and Mixture of Attribute Experts (MoAE). First, MoPE is inserted into the transformer using a dual-level routing mechanism to conditionally dispatch image patches to specialized experts. This ensures coherent attribute families are processed by dedicated experts. Finally, the MoAE head projects these expert-refined features into sparse, partaware attribute maps for robust zero-shot classification. On zero-shot learning benchmark datasets CUB, AwA2, and SUN, our ACR achieves consistent state-of-the-art results.

An approach of deep reinforcement learning for maximizing the net present value of stochastic projects

This paper investigates a project with stochastic activity durations and cash flows under discrete scenarios, where activities must satisfy precedence constraints generating cash inflows and outflows. The objective is to maximize expected net present value (NPV) by accelerating inflows and deferring outflows. We formulate the problem as a discrete-time Markov Decision Process (MDP) and propose a Double Deep Q-Network (DDQN) approach. Comparative experiments demonstrate that DDQN outperforms traditional rigid and dynamic strategies, particularly in large-scale or highly uncertain environments, exhibiting superior computational capability, policy reliability, and adaptability. Ablation studies further reveal that the dual-network architecture mitigates overestimation of action values, while the target network substantially improves training convergence and robustness. These results indicate that DDQN not only achieves higher expected NPV in complex project optimization but also provides a reliable framework for stable and effective policy implementation.

Virtual Width Networks

We introduce Virtual Width Networks (VWN), a framework that delivers the benefits of wider representations without incurring the quadratic cost of increasing the hidden size. VWN decouples representational width from backbone width, expanding the embedding space while keeping backbone compute nearly constant. In our large-scale experiment, an 8-times expansion accelerates optimization by over 2 times for next-token and 3 times for next-2-token prediction. The advantage amplifies over training as both the loss gap grows and the convergence-speedup ratio increases, showing that VWN is not only token-efficient but also increasingly effective with scale. Moreover, we identify an approximately log-linear scaling relation between virtual width and loss reduction, offering an initial empirical basis and motivation for exploring virtual-width scaling as a new dimension of large-model efficiency.

Distributed Zero-Shot Learning for Visual Recognition

In this paper, we propose a Distributed Zero-Shot Learning (DistZSL) framework that can fully exploit decentralized data to learn an effective model for unseen classes. Considering the data heterogeneity issues across distributed nodes, we introduce two key components to ensure the effective learning of DistZSL: a cross-node attribute regularizer and a global attribute-to-visual consensus. Our proposed cross-node attribute regularizer enforces the distances between attribute features to be similar across different nodes. In this manner, the overall attribute feature space would be stable during learning, and thus facilitate the establishment of visual-to-attribute(V2A) relationships. Then, we introduce the global attribute-tovisual consensus to mitigate biased V2A mappings learned from individual nodes. Specifically, we enforce the bilateral mapping between the attribute and visual feature distributions to be consistent across different nodes. Thus, the learned consistent V2A mapping can significantly enhance zero-shot learning across different nodes. Extensive experiments demonstrate that DistZSL achieves superior performance to the state-of-the-art in learning from distributed data.

The Universal Landscape of Human Reasoning

Understanding how information is dynamically accumulated and transformed in human reasoning has long challenged cognitive psychology, philosophy, and artificial intelligence. Existing accounts, from classical logic to probabilistic models, illuminate aspects of output or individual modelling, but do not offer a unified, quantitative description of general human reasoning dynamics. To solve this, we introduce Information Flow Tracking (IF-Track), that uses large language models (LLMs) as probabilistic encoder to quantify information entropy and gain at each reasoning step. Through fine-grained analyses across diverse tasks, our method is the first successfully models the universal landscape of human reasoning behaviors within a single metric space. We show that IF-Track captures essential reasoning features, identifies systematic error patterns, and characterizes individual differences. Applied to discussion of advanced psychological theory, we first reconcile single- versus dual-process theories in IF-Track and discover the alignment of artificial and human cognition and how LLMs reshaping human reasoning process. This approach establishes a quantitative bridge between theory and measurement, offering mechanistic insights into the architecture of reasoning.

CoRA: Covariate-Aware Adaptation of Time Series Foundation Models

Time Series Foundation Models (TSFMs) have shown significant impact through their model capacity, scalability, and zero-shot generalization. However, due to the heterogeneity of inter-variate dependencies and the backbone scalability on large-scale multivariate datasets, most TSFMs are typically pre-trained on univariate time series. This limitation renders them oblivious to crucial information from diverse covariates in real-world forecasting tasks. To further enhance the performance of TSFMs, we propose a general covariate-aware adaptation (CoRA) framework for TSFMs. It leverages pre-trained backbones of foundation models while effectively incorporating exogenous covariates from various modalities, including time series, language, and images, to improve the quality of predictions. Technically, CoRA maintains the equivalence of initialization and parameter consistency during adaptation. With preserved backbones of foundation models as frozen feature extractors, the outcome embeddings from foundation models are empirically demonstrated more informative than raw data. Further, CoRA employs a novel Granger Causality Embedding (GCE) to automatically evaluate covariates regarding their causal predictability with respect to the target variate. We incorporate these weighted embeddings with a zero-initialized condition-injection mechanism, avoiding catastrophic forgetting of pre-trained foundation models and gradually integrates exogenous information. Extensive experiments show that CoRA of TSFMs surpasses state-of-the-art covariate-aware deep forecasters with full or few-shot training samples, achieving 31.1% MSE reduction on covariate-aware forecasting. Compared to other adaptation methods, CoRA exhibits strong compatibility with various advanced TSFMs and extends the scope of covariates to other modalities, presenting a practical paradigm for the application of TSFMs.