Search papers, labs, and topics across Lattice.
Affiliation:
6
0
11
An analytical framework for power-aware provisioning of PD-disaggregated AI inference is developed, derived from the joint distribution of input--output lengths and hardware compute and memory limits, which determines the serving capacity and average power consumption of a provisioned deployment.
EPIC slashes memory usage by 27.5x and energy consumption by 24.3x, making intelligent AR glasses feasible for everyday use.
Speculative reasoning in multimodal models gets a serious upgrade: DREAM-R achieves significant speedups without compromising accuracy by aligning speculative drafts with target reasoning trajectories via RL.
Current egocentric video benchmarks miss the mark: EgoEverything uses human gaze to create questions that actually reflect how people behave, not just what they see.
Personalized federated learning can boost VLN performance by up to 7.8% in trajectory fidelity and converge 1.38x faster, by selectively fusing parameters in environment-sensitive layers.
