0:00 / 0:00
research
Research Jam #29
ML Collective16 June 2026Watch on YouTube
Part of series
Ep. 2 · Jam Research
View the seriesDescription
MLC: Open Collab is a 100% open community for independent researchers. We held our twenty ninth Research Jam on October 1, 2025, where presenters signed up to share updates on their ongoing research. Access slides and read more about it at https://mlcollective.org/events/research-jam-29/ 00:24 Justin Jung: Interactive demo of 3D minecraft structure generation 11:54 Atharva Bhutani: Compute-Accuracy Tradeoff in N-bit Neural Networks 25:07 Bryce Sandlund: Improving sample efficiency in LLM RL 34:34 Raymond Fan: How Data Affects Loss Landscape Volumes
What you'll learn
- Three research themes take center stage: 3D structure generation with neural networks, model efficiency optimization, and improvements to training processes
- Quantization and data quality play essential roles in balancing computational power against model accuracy
- Open collaboration among independent researchers drives innovations in deep learning and reinforcement learning applications
Frequently asked questions
What four research projects were presented at Research Jam #29?
The presentations covered interactive 3D Minecraft structure generation, compute-accuracy tradeoffs in N-bit neural networks, improving sample efficiency in LLM reinforcement learning, and how data affects loss landscape volumes.
What is the purpose of ML Collective's Research Jam events?
Research Jam is a platform where independent researchers can share updates on their ongoing projects within ML Collective's 100% open community dedicated to neural networks and AI research.
How do quantization and data quality relate to each other in AI training?
Both topics address fundamental efficiency questions: quantization optimizes computational cost versus accuracy, while data quality determines how well models train and generalize.