Episodes (Page 11)
✨
Compares GRPO and DPO reinforcement learning algorithms for text-to-image generation.
✨
Presents Let Androids Dream (LAD) framework for understanding implied meanings in images.
✨
Introduces SmolVLM, small-scale multimodal models for efficient computing.
✨
Surveys Federated Learning (FL), a distributed approach for collaborative training.
✨
Focuses on efficient Federated Learning for autonomous mobile networks using Tiny Language Models.
✨
Introduces Mobile-MMLU benchmark for evaluating LLMs on mobile devices.
✨
Presents AI-RAN, unifying Radio Access Network and AI workloads.
✨
Explains fine-tuning LLMs for specialized tasks, not for new factual knowledge.
✨
Describes an LLM customized for explaining VHDL code in processor design.
✨
Introduced Adaptively Weighted Nearest Neighbors (AWNN) for matrix completion.
✨
Explored gradient flow dynamics in neural networks.
✨
Introduced WavReward for evaluating end-to-end spoken dialogue models.
✨
Introduced BLIP3-o, a unified multimodal model for image understanding and generation.
✨
Introduced CodePDE, using LLMs to generate PDE solver code.
✨
Investigated online learning for feedforward neural networks with sign activation.
✨
Introduced CityAVOS, a benchmark dataset for UAV visual object search in urban areas.
✨
Introduced BAT, a benchmark for auto-bidding algorithms in online advertising.
✨
Highlighted Amazon's AI, ML, and robotics research.
Human Feedback Improvements
✨
Introduced T2I-R1 for text-to-image generation using RL and bi-level CoT.
✨
Proposes pretraining for estimating heterogeneous treatment effects (HTE).