FutureMLS Lab

The Future Machine Learning & Systems Lab is a nonprofit research lab working where machine learning meets computer systems. Foundation models are remarkably capable but costly to train and serve. Our mission is to close the gap between model capability and the real-world cost of deploying it.

We pursue algorithm–system co-design in four connected directions:

• Efficient ML architecture: speculator and attention design
• Efficient post-training: loss design, reinforcement learning and quantization
• Efficient post-training engines: post-training and inference systems
• Agents: coding agents and AI for science

Our work spans methods, models, kernels, runtimes and serving systems. It is open source, reproducible and built to be used.

Selected projects: OSCAR, 2-bit KV-cache quantization (NeurIPS 2026) · CARE, covariance-aware conversion to Multi-Head Latent Attention (ICLR 2026) · Taylor-Calibrate, hybrid linear-attention distillation

Founder & PI: Zhongzhu (Charlie) Zhou · Website · GitHub · zhongzhu@futuremls.org