Abstract--Large language models (LLMs) offer broad utility but remain prone to hallucination and out-of-distribution (OOD) errors. We propose EigenTrack, an interpretable real-time detector that uses the spectral geometry of hidden activations, a compact global signature of model dynamics. By streaming covariance-spectrum statistics such as entropy, eigenvalue gaps, and KL divergence from random baselines into a lightweight recurrent classifier, EigenTrack tracks temporal shifts in representation structure that signal hallucination and OOD drift before surface errors appear . Unlike black-and grey-box methods, it needs only a single forward pass without resampling. Unlike existing white-box detectors, it preserves temporal context, aggregates global signals, and offers interpretable accuracy-latency trade-offs.

Large deep learning models such as BERT and ResNet achieve state-of-the-art performance but are costly to deploy at the edge due to their size and compute demands. We present RMT-KD, a compression method that leverages Random Matrix Theory (RMT) for knowledge distillation to iteratively reduce network size. Instead of pruning or heuristic rank selection, RMT-KD preserves only informative directions identified via the spectral properties of hidden representations. RMT-based causal reduction is applied layer by layer with self-distillation to maintain stability and accuracy. On GLUE, AG News, and CIFAR-10, RMT-KD achieves up to 80% parameter reduction with only 2% accuracy loss, delivering 2.8x faster inference and nearly halved power consumption. These results establish RMT-KD as a mathematically grounded approach to network distillation.

Reward models are central to both reinforcement learning (RL) with language models and inference-time verification. However, existing reward models often lack temporal consistency, leading to ineffective policy updates and unstable RL training. We introduce TDRM, a method for learning smoother and more reliable reward models by minimizing temporal differences (TD) for training-time reinforcement learning and inference-time verification. Experiments show that TD-trained process reward models (PRMs) improve performance across Best-of-N (up to 6.6%) and tree-search (up to 23.7%) settings. When combined with Reinforcement Learning with Verifiable Rewards (RLVR), TD-trained PRMs lead to more data-efficient RL -- achieving comparable performance with just 2.5k data to what baseline methods require 50.1k data to attain -- and yield higher-quality language model policies in 8 model variants (5 series), e.g., Qwen2.5-(0.5B, 1,5B), GLM4-9B-0414, GLM-Z1-9B-0414, Qwen2.5-Math-(1.5B, 7B), and DeepSeek-R1-Distill-Qwen-(1.5B, 7B). We release all code at https://github.com/THUDM/TDRM.

Generative artificial intelligence (GenAI) tools such as OpenAI's ChatGPT are transforming the educational landscape, prompting reconsideration of traditional assessment practices. In parallel, universities are exploring alternatives to in-person, closed-book examinations, raising concerns about academic integrity and pedagogical alignment in uninvigilated settings. This study investigates whether traditional closed-book mathematics examinations retain their pedagogical relevance when hypothetically administered in uninvigilated, open-book settings with GenAI access. Adopting an empirical approach, we generate, transcribe, and blind-mark GenAI submissions to eight undergraduate mathematics examinations at a Russell Group university, spanning the entirety of the first-year curriculum. By combining independent GenAI responses to individual questions, we enable a meaningful evaluation of GenAI performance, both at the level of modules and across the first-year curriculum. We find that GenAI attainment is at the level of a first-class degree, though current performance can vary between modules. Further, we find that GenAI performance is remarkably consistent when viewed across the entire curriculum, significantly more so than that of students in invigilated examinations. Our findings evidence the need for redesigning assessments in mathematics for unsupervised settings, and highlight the potential reduction in pedagogical value of current standards in the era of generative artificial intelligence.

Physical commonsense reasoning datasets like PIQA are predominantly English-centric and lack cultural diversity. We introduce Ko-PIQA, a Korean physical commonsense reasoning dataset that incorporates cultural context. Starting from 3.01 million web-crawled questions, we employed a multi-stage filtering approach using three language models to identify 11,553 PIQA-style questions. Through GPT-4o refinement and human validation, we obtained 441 high-quality question-answer pairs. A key feature of Ko-PIQA is its cultural grounding: 19.7% of questions contain culturally specific elements like traditional Korean foods (kimchi), clothing (hanbok), and specialized appliances (kimchi refrigerators) that require culturally-aware reasoning beyond direct translation. We evaluate seven language models on Ko-PIQA, with the best model achieving 83.22% accuracy while the weakest reaches only 59.86%, demonstrating significant room for improvement. Models particularly struggle with culturally specific scenarios, highlighting the importance of culturally diverse datasets. Ko-PIQA serves as both a benchmark for Korean language models and a foundation for more inclusive commonsense reasoning research. The dataset and code will be publicly available.

Large Foundation Models (LFMs) have demonstrated significant advantages in civil engineering, but they primarily focus on textual and visual data, overlooking the rich semantic, spatial, and topological features in BIM (Building Information Modelling) models. Therefore, this study develops the first large-scale graph neural network (GNN), BIGNet, to learn, and reuse multidimensional design features embedded in BIM models. Firstly, a scalable graph representation is introduced to encode the "semantic-spatial-topological" features of BIM components, and a dataset with nearly 1 million nodes and 3.5 million edges is created. Subsequently, BIGNet is proposed by introducing a new message-passing mechanism to GraphMAE2 and further pretrained with a node masking strategy. Finally, BIGNet is evaluated in various transfer learning tasks for BIM-based design checking. Results show that: 1) homogeneous graph representation outperforms heterogeneous graph in learning design features, 2) considering local spatial relationships in a 30 cm radius enhances performance, and 3) BIGNet with GAT (Graph Attention Network)-based feature extraction achieves the best transfer learning results. This innovation leads to a 72.7% improvement in Average F1-score over non-pretrained models, demonstrating its effectiveness in learning and transferring BIM design features and facilitating their automated application in future design and lifecycle management.

We introduce Delayed Streams Modeling (DSM), a flexible formulation for streaming, multimodal sequence-to-sequence learning. Sequence-to-sequence generation is often cast in an offline manner, where the model consumes the complete input sequence before generating the first output timestep. Alternatively, streaming sequence-to-sequence rely on learning a policy for choosing when to advance on the input stream, or write to the output stream. DSM instead models already time-aligned streams with a decoder-only language model. By moving the alignment to a pre-processing step,and introducing appropriate delays between streams, DSM provides streaming inference of arbitrary output sequences, from any input combination, making it applicable to many sequence-to-sequence problems. In particular, given text and audio streams, automatic speech recognition (ASR) corresponds to the text stream being delayed, while the opposite gives a text-to-speech (TTS) model. We perform extensive experiments for these two major sequence-to-sequence tasks, showing that DSM provides state-of-the-art performance and latency while supporting arbitrary long sequences, being even competitive with offline baselines. Code, samples and demos are available at https://github.com/kyutai-labs/delayed-streams-modeling

Multispectral object detection is an important application for unmanned aerial vehicles (UAVs). However, it faces several challenges. First, low-light RGB images weaken the multispectral fusion due to details loss. Second, the interference information is introduced to local target modeling during multispectral fusion. Third, computational cost poses deployment challenge on UAV platforms, such as transformer-based methods with quadratic complexity. To address these issues, a framework named DEPFusion consisting of two designed modules, Dual-Domain Enhancement (DDE) and Priority-Guided Mamba Fusion (PGMF) , is proposed for UAV multispectral object detection. Firstly, considering the adoption of low-frequency component for global brightness enhancement and frequency spectra features for texture-details recovery, DDE module is designed with Cross-Scale Wavelet Mamba (CSWM) block and Fourier Details Recovery (FDR) block. Secondly, considering guiding the scanning of Mamba from high priority score tokens, which contain local target feature, a novel Priority-Guided Serialization is proposed with theoretical proof. Based on it, PGMF module is designed for multispectral feature fusion, which enhance local modeling and reduce interference information. Experiments on DroneVehicle and VEDAI datasets demonstrate that DEPFusion achieves good performance with state-of-the-art methods.

In this paper, a green learning (GL)-based precoding framework is proposed for simultaneously transmitting and reflecting reconfigurable intelligent surface (STAR-RIS)-aided millimeter-wave (mmWave) MIMO broadcasting systems. Motivated by the growing emphasis on environmental sustainability in future 6G networks, this work adopts a broadcasting transmission architecture for scenarios where multiple users share identical information, improving spectral efficiency and reducing redundant transmissions and power consumption. Different from conventional optimization methods, such as block coordinate descent (BCD) that require perfect channel state information (CSI) and iterative computation, the proposed GL framework operates directly on received uplink pilot signals without explicit CSI estimation. Unlike deep learning (DL) approaches that require CSI-based labels for training, the proposed GL approach also avoids deep neural networks and backpropagation, leading to a more lightweight design. Although the proposed GL framework is trained with supervision generated by BCD under full CSI, inference is performed in a fully CSI-free manner. The proposed GL integrates subspace approximation with adjusted bias (Saab), relevant feature test (RFT)-based supervised feature selection, and eXtreme gradient boosting (XGBoost)-based decision learning to jointly predict the STAR-RIS coefficients and transmit precoder. Simulation results show that the proposed GL approach achieves competitive spectral efficiency compared to BCD and DL-based models, while reducing floating-point operations (FLOPs) by over four orders of magnitude. These advantages make the proposed GL approach highly suitable for real-time deployment in energy- and hardware-constrained broadcasting scenarios.

Modern e-commerce platforms strive to enhance customer experience by providing timely and contextually relevant recommendations. However, recommending general merchandise to customers focused on grocery shopping -- such as pairing milk with a milk frother -- remains a critical yet under-explored challenge. This paper introduces a cross-pollination (XP) framework, a novel approach that bridges grocery and general merchandise cross-category recommendations by leveraging multi-source product associations and real-time cart context. Our solution employs a two-stage framework: (1) A candidate generation mechanism that uses co-purchase market basket analysis and LLM-based approach to identify novel item-item associations; and (2) a transformer-based ranker that leverages the real-time sequential cart context and optimizes for engagement signals such as add-to-carts. Offline analysis and online A/B tests show an increase of 36\% add-to-cart rate with LLM-based retrieval on the item page, and 15\% lift in add-to-cart using cart context-based ranker on the cart page. Our work contributes practical techniques for cross-category recommendations and broader insights for e-commerce systems.