Peng Jiang

RecHarness boosts recommender system performance by over 2% in real-world applications while minimizing the need for manual engineering interventions.

LGRID transforms the landscape of local-life service recommendations by generating interpretable Semantic IDs that are both semantically rich and collision-resistant.

Latent reasoning can boost recommendation efficiency by over 10x while enhancing accuracy, challenging the need for verbose rationales in LLM applications.

Generating personalized videos on demand can boost ad revenue by nearly 2% in high-traffic environments, transforming traditional recommendation systems.

Aligning LLM reasoning with a dedicated recommendation head via reinforcement learning yields state-of-the-art recommendation performance in real-world systems.

LLMs can generate recommendations up to 3.1x faster by explicitly modeling token position within items and speculation depth during speculative decoding.

Radar odometry, typically confined to urban settings, can be pushed off-road with simple adaptations like IMU preintegration, but still faces significant challenges in unstructured environments.

Two-tower recommendation models can get a major online performance boost without latency penalties, thanks to a new capability synergy framework.

Forget disjointed workflows: AutoCut's unified token space for video, audio, and text slashes ad production costs while boosting consistency.

Generative recommendation can beat DLRM in large-scale advertising, driving a 4.2% revenue lift in Kuaishou's production system via innovations in tokenization, decoding, optimization, and serving.

Ditch rigid distribution assumptions: a novel residual quantization approach predicts continuous values by recursively refining quantization codes, outperforming SOTA in recommendation tasks.