AI Research

Researchers propose a memory-augmented LSTM autoencoder for unsupervised human activity recognition using IMU sensor fusion, achieving 96.6% and 98.4% accuracy on DaLiAc and PAMAP2 datasets, outperforming supervised base…

Researchers trained a Joint Embedding Predictive Architecture (JEPA) on unlabeled driving data from the nuPlan dataset and used temporal prediction error as a zero-shot complexity score, finding it effectively identifies…

Researchers propose a variation of Fidelity-based XAI metrics for evaluating explanations of CNN classifiers in real-world applications with few classes. The method generates in-distribution, uncertainty-provoking pertur…

Researchers introduced RADIANT-PET, a framework combining PET/CT lesion segmentation with large language model reasoning and reinforcement learning, which outperformed image-only baselines in reducing false positives and…

Researchers introduced Jenga Inverse Predictor (JIP-2), a GPU-accelerated deep learning framework that reconstructs collapsed stone structures by treating anastylosis as an inverse prediction task. The system uses a phys…

Researchers introduced JASPR, a self-supervised deep learning framework that integrates histology images and spatial transcriptomics data to improve virtual genomic screening and clinical prognostication. Trained on brea…

Researchers propose a semantic-aware generative image transmission framework for resource-constrained visual IoT systems. The method selects and transmits only task-relevant image tokens based on semantic importance, ach…

Researchers propose ViPSy, a framework for constructing preference data that reduces hallucinations in Vision-Language Models (VLMs) by leveraging visual cues from semantically aligned image variants. The method achieves…

Researchers at the University of Oxford and other institutions found that large language models (LLMs) develop false-belief task (FBT) performance late in pretraining, dependent on model size and training volume, and imp…

Researchers introduced a French OSCE dialogue dataset with 240 student-patient interactions and developed a controllable LLM-based pipeline to generate synthetic OSCE dialogues for clinical training. The system includes …

Researchers introduced turn-averaged sparse autoencoders (SAEs) that represent entire conversational turns with fixed feature sets, improving interpretability of long-context language models. The method outperforms per-t…

Researchers at arXiv show that static per-layer Fibonacci spacing in sparse attention outperforms learned dilation and fixed schedules across 21 language models, achieving better perplexity and extrapolating to four time…

Researchers propose grain calibration to validate whether large language models measure theoretical constructs correctly, not just reliably. The method decomposes constructs into clause-level components and tests each wi…

Researchers introduced SEATauBench, the first agent-focused evaluation framework for Southeast Asian sovereign AI, adapting TauBench to five languages. Testing across three models revealed that English agent capabilities…

Researchers introduced 5ting, a system for the SemEval-2026 Task 8 (MTRAGEval) that combines BGE-M3 dense retrieval, FAISS indexing, dual-query merged retrieval, and LLM-based reranking with role-separated generation con…

Researchers developed a structure-preserving Marathi-to-English document translation framework using a multi-stage LLM pipeline, integrating layout-aware OCR and HTML reconstruction to maintain formatting and domain term…

Researchers at arXiv found that using large language models (LLMs) like GPT-5.2 to label training data for entity matching can reduce manual labeling effort by hundreds of hours, with student models performing within two…

Researchers prove that mesh intelligence requires a liquid substrate—a continuous-time, multi-timescale network—because agents must estimate a changing latent from irregular, unscheduled observations without retraining. …

Reinforcement learning researchers need to clearly distinguish between using simulators as a proxy for real-world deployment and solving simulators as an end in itself, according to a new paper. The authors argue that co…

Researchers propose S-GAI, a spectral geometry-aware initialization framework for sigmoidal MLPs that encodes dataset geometry into network weights. Using SVD to extract class-wise spectral features, S-GAI initializes hi…