Yuan Cheng

The DeepSeek-V4.1-Flash model, a multimodal Mixture-of-Experts model with 552B backbone parameters and support for contexts of up to one million tokens, is introduced, substantially improving cost efficiency for agentic workloads and pushing the limits of KV cache compression.

This work proposes UniRec, a Unified Cross-stage Recommendation Fusion model, and introduces a dual-axis preference alignment objective, and finds that unconstrained end-to-end fusion optimization can exploit imbalances in item attribute distributions, over-concentrating on high-reward regions at the cost of other objectives.

A groundbreaking dataset and benchmark that bridges the gap between anatomical and metabolic analysis in whole-body PET/CT imaging, enabling advanced multimodal reasoning.

Predicting the next token's KV entries can boost long-context LLM throughput by over 2.5 times without sacrificing latency or quality.

PIVOT achieves up to 4x faster indexing for token-level sparse attention without sacrificing accuracy, transforming how we handle query processing in large models.

Exact Aumann-Shapley attributions in GNNs are now achievable with a polynomial architecture, drastically improving fidelity while slashing computational costs.

Only a subset of design interactions in heterogeneous LLM inference are binding constraints, revealing critical insights for optimizing deployment strategies.