Jiaqi Zhang

LSA slashes indexing overhead while maintaining full attention performance, enabling efficient long-context processing for models with up to one million tokens.

Zhinv slashes wind field reconstruction errors by 66%, enabling real-time assessments directly from local observations.

Hard interventions can reveal causal structures even when traditional assumptions of faithfulness fail, challenging the status quo in causal discovery.

Cross-component filters can dramatically enhance chroma fidelity by exploiting luma-chroma correlations, paving the way for superior video quality.

By synthesizing diverse training data and leveraging pixel-wise calibration, FoundationGeo achieves unprecedented robustness in monocular metric geometry, even under varying camera intrinsics.

Long-context LLM rankings dramatically reshuffle when evaluated across a range of context lengths and capabilities, proving that a single headline score is misleading.

MONA unlocks faster LLM pretraining and superior downstream performance by turbocharging the Muon optimizer with Nesterov-style acceleration, leaving AdamW in the dust.

Quadrupedal robots can now perform dynamic loco-manipulation in the real world, matching human teleoperation, using only onboard ego-centric vision and a low-frequency (5Hz) open-vocabulary detector.

Achieve full-attention accuracy with 10x operator speedup and 4.7x throughput improvement in long-context LLM inference by overlapping KV cache transfers with computation.