Senior Machine Learning Research Coauthor — VAE Representation Learning for Latent Diffusion

Bradbury Group is a volunteer-run, distributed AI research lab developing novel methods in image diffusion, scientific machine learning, and lightweight generative models. Supported by partners such as W&B, Lambda Labs, and Modal, we run large-scale experiments and publish work aimed at top ML venues.

We are assembling a large, high-skill research team for a new paper on representation learning and VAE design for latent diffusion models.

Project details are intentionally abstracted but involve designing and evaluating next-generation latent spaces for controllable diffusion.

What you will work on:

• Designing, implementing, and iterating a novel VAE architecture for image generative modeling.
• Exploring how latent complexity (and other factors) influence downstream diffusion performance and controllability.
• Running large-scale experiments, diagnostics, and ablations to evaluate latent geometry.
• Contributing to a publication-grade research pipeline with senior researchers.
• (Further project details intentionally withheld.)

(The requirements for this project are steeper than most of our others. If you do not meet these criteria please check our other roles.)

Must have:

• First-author paper at a top ML venue (NeurIPS, ICLR, ICML, CVPR, etc.).
• Deep experience training diffusion models end-to-end.
• Strong VAE experience; for example VAE, VQ-VAE, VQ-VAE-GAN (architectures + training dynamics).
• Exceptional PyTorch expertise, comfortable modifying large models and writing custom training loops.
• Exceptional deep neural network design skills, and knowledge of scaling laws etc.
• Demonstrated record of self-directed research (PhD, postdoc, or substantial lab work resulting in publications).

Preferred:

• Pretraining experience with large diffusion or language models.
• Experience training on web-scale datasets.
• Strong proficiency with DDP, FSDP, and large-batch training.
• Solid mathematical foundations in optimization, representation learning, or generative modeling.

This role is best suited for industry research engineers, postdocs, or highly experienced PhD-level researchers. Please refer people who may be a good fit.

What we offer:

• Work alongside senior researchers in a highly parallelized, ambitious project targeting a major publication.
• Access to multi-GPU H100 clusters through our compute sponsors.
• A fast, clean research environment — no bureaucracy, no overhead, no corporate constraints.
• Full co-author status, with first-author possibilities depending on contribution (authorship tracked fairly through mutual agreement).
• Opportunity to play a key role in shaping the lab and contribute to other ongoing Bradbury Group projects (optional).

This is a volunteer research role, not a salaried job. Our researchers come from top institutions and labs; expectations are seriousness, creativity, and strong scientific discipline.

We strongly encourage including a cover letter or personal note via LinkedIn or our website contact form describing your background, research experience, and interest in this project.