6238 papers
ELISA is an interpretable hybrid AI agent that bridges natural language and single-cell transcriptomic data by integrating scGPT embeddings with LLM reasoning to enable interactive, expression-grounded biological discovery, outperforming existing tools in cell type retrieval and hypothesis generation across diverse datasets.
AdaFuse is a framework that accelerates dynamic adapter inference in Large Language Models by employing a token-level pre-gating strategy to enable a single global routing decision, which is then executed via a custom fused CUDA kernel to reduce decoding latency by over 2.4x while maintaining accuracy.
The paper introduces "Mirror," a data-curation design pattern that utilizes a strictly curated 32-cell topology and a lightweight linear SVM to achieve superior speed, determinism, and detection accuracy for prompt injection screening compared to large neural models, demonstrating that strict data geometry is more critical than model scale for initial defense layers.
This paper introduces Bielik-Minitron-7B, a compressed 7.35B-parameter Polish language model created by applying structured pruning and knowledge distillation to the Bielik-11B-v3.0 model, which achieves a 33.4% parameter reduction and up to 50% inference speedup while retaining approximately 90% of the original model's performance.
The paper proposes "Think While Watching," a memory-anchored streaming framework that enables efficient multi-turn video reasoning in multimodal large language models by preserving segment-level memory and overlapping perception with generation, thereby significantly improving accuracy on streaming benchmarks while reducing output tokens.
EnTransformer is a deep generative framework that combines the sequence modeling power of Transformers with the stochastic learning paradigm of engression to produce well-calibrated, non-parametric multivariate probabilistic forecasts that outperform existing benchmarks.
This paper systematically evaluates how nine large language models, including the latest versions, often fail to refuse processing user-supplied harmful content even when performing benign tasks, revealing a critical content-level ethical vulnerability that current safety alignment overlooks.
This paper introduces MobileKernelBench, a framework revealing that current LLMs struggle to generate efficient mobile kernels due to compilation failures and hallucinations, and proposes MoKA, a multi-agent system that significantly improves compilation success and kernel performance.
This paper introduces Fair-PaperRec, an MLP-based model that effectively mitigates demographic biases in paper recommendations by penalizing disparities through intersectional criteria and a customized fairness loss, achieving a significant increase in underrepresented group participation while simultaneously improving overall utility without compromising academic rigor.
The paper proposes ProtoSR, a prototype-based framework that leverages an instruction-tuned LLM to extract visual prototypes from free-text radiology reports, thereby injecting unstructured knowledge to significantly improve fine-grained structured report generation and achieve state-of-the-art performance on the Rad-ReStruct benchmark.
This paper introduces Effective Resistance Rewiring (ERR), a parameter-free topological correction method that iteratively optimizes graph connectivity by adding and removing edges based on global effective resistance to alleviate over-squashing in Graph Neural Networks, while demonstrating that combining this approach with normalization techniques effectively balances the trade-off between improved long-range signal propagation and oversmoothing.
This paper proposes integrating Voronoi tessellations into probabilistic circuits to capture local data geometry, addressing the resulting tractability challenges through both an approximate inference framework with guaranteed bounds and a structural condition for exact inference, while enabling gradient-based learning via a differentiable relaxation.
This paper introduces Delayed Backdoor Attacks (DBA), a novel threat paradigm that decouples trigger exposure from malicious activation via a temporal dimension, enabling the use of common words as triggers and demonstrating the feasibility of the DND prototype which remains dormant before achieving near-perfect attack success rates while evading current defenses.
This paper introduces SLIP, an open-source framework that leverages language-informed pretraining with a flexible patch-embedder and cross-attention mechanism to learn transferable sensor representations capable of handling diverse configurations and achieving superior zero-shot performance in classification, captioning, and question answering across 11 datasets.
This paper proposes a multimodal emotion recognition framework for the ABAW 10th EXPR challenge that combines frozen CLIP and Wav2Vec 2.0 backbones with temporal convolutional networks and a bi-directional cross-attention fusion module to achieve robust performance in unconstrained real-world environments.
This paper introduces NormCoRe, a novel methodological framework that systematically translates human subject experiments into multi-agent AI environments to study normative coordination, demonstrating through a distributive justice replication that AI agents' normative judgments differ from human baselines and are sensitive to foundation model and persona choices.
This paper introduces HomeSafe-Bench, a comprehensive benchmark for evaluating unsafe action detection in household scenarios using 438 diverse cases, and proposes HD-Guard, a hierarchical dual-brain architecture that effectively balances real-time inference efficiency with deep multimodal reasoning safety monitoring.
This paper introduces LABSHIELD, a multimodal benchmark grounded in OSHA and GHS standards that evaluates the safety awareness and reasoning capabilities of large language models in laboratory settings, revealing a significant performance gap in hazard identification and safety-critical planning compared to general-domain tasks.
This paper introduces BTZSC, a comprehensive benchmark of 22 datasets designed to systematically evaluate and compare the zero-shot text classification capabilities of NLI cross-encoders, embedding models, rerankers, and instruction-tuned LLMs, revealing that modern rerankers currently achieve state-of-the-art performance while embedding models offer the best accuracy-latency trade-off.
This paper introduces FedFew, a personalized federated learning framework that reformulates the problem as a few-for-many optimization task to efficiently serve thousands of heterogeneous clients with only a few shared server models, achieving near-optimal personalization without manual clustering or hyperparameter tuning.