Search papers, labs, and topics across Lattice.
13 papers from Google Research on Training Efficiency & Optimization
Refining generative models with discriminator guidance provably improves generalization, offering a theoretical justification for techniques like score-based diffusion.
Forget end-to-end training: DexMulti's "retrieve-align-execute" approach lets robots master complex, multi-stage dexterous tasks from just a handful of demonstrations.
Dataset condensation, previously limited to neural networks, can now democratize access to clinical data by enabling privacy-preserving training of classical models like decision trees and Cox regression.
Forget catastrophic forgetting: this function-preserving expansion method lets you fine-tune without sacrificing pre-trained knowledge, matching full fine-tuning performance at a fraction of the cost.
See in the dark: Dark3R unlocks structure from motion at signal-to-noise ratios below -4dB, where existing methods completely break down.
Forget quadratic scaling: ZipMap zips entire 3D scenes from hundreds of images into a compact state in a single pass, unlocking 20x faster reconstruction.
DARKFormer closes the performance gap with exact softmax attention in finetuning by learning a data-aligned kernel geometry for efficient random feature approximation, sidestepping the need for retraining or large feature budgets.
Recurrent models can now achieve Transformer-competitive performance on recall-intensive tasks, thanks to a simple memory caching mechanism that grows memory capacity with sequence length.
Ditch slow, external segmentation pipelines: TrajTok learns trajectory tokens end-to-end, boosting video understanding while staying lean and adaptable.
Forget fine-tuning: Prompt-Level Distillation lets small models match frontier reasoning performance by distilling explicit reasoning patterns into structured system prompts.
Surprisingly, using only a single inner loop update in data mixing can lead to failure, and the optimal number of inner loop steps scales logarithmically with the parameter update budget.
Ditch Stable Diffusion's latents: Unified Latents (UL) achieves state-of-the-art video generation and competitive image generation with fewer training FLOPs.
Randomly masking parameter updates in RMSProp delivers state-of-the-art LLM training performance, revealing a surprisingly effective form of geometric regularization.
