JB.

Jarred Barber

Machine Learning Research Engineer — Cambridge, MA

Email GitHub Scholar LinkedIn Print / PDF
Experience
Staff Research Engineer 2022 — Present
Google DeepMind / Gemini Thinking Team (Cambridge, MA)
  • Tech lead on latent reasoning and pretraining research for the Gemini models.
  • Developed custom attention kernels in JAX/Pallas for efficient transformer training.
  • Designed agent workflows and harnesses for mathematical and code reasoning.
  • Led the Muse project (ICML 2023): delivered Google's first generative image product; the novel editing algorithm was featured at the Google I/O 2023 keynote and launched on Vertex AI.
  • Mentored two MIT PhD researchers; inventor on four filed patents.
Applied Scientist, Alexa AI 2020 — 2022
Amazon / Speech Recognition
  • Led a five-person team on robust ASR for challenging acoustic environments.
  • Replaced legacy heuristic systems with learned ML models.
  • Researched generative audio models: VQ-VAE, normalizing flows, diffusion.
Machine Learning Scientist 2019 — 2020
Charles River Analytics / DARPA Programs
  • Tech lead for three DARPA AI research programs on generative models and meta-learning.
ML Research Scientist 2017 — 2019
FeatureX / Acquired by Orbital Insight
  • Computer vision tech lead at a satellite imagery startup (acquired Sept 2018).
  • Wrote optimized CUDA kernels for differentiable resampling algorithms.
Associate Technical Staff 2011 — 2017
MIT Lincoln Laboratory / Airborne Radar Systems
  • PI / tech lead on ML and signal processing research for imaging radar systems.
Software Engineer II 2008 — 2011
Lockheed Martin
  • Tech lead on ML models for intelligence analysis applications.
Selected publications
2025
Gemini 2.5: Pushing the frontier with advanced reasoning, multimodality, long context, and next generation agentic capabilities
Gemini Team
arXiv 3,400+ cites
2024
StyleDrop: Text-to-image synthesis of any style
K. Sohn, L. Jiang, J. Barber, et al.
NeurIPS 310 cites
2023
Muse: Text-to-image generation via masked generative transformers
H. Chang, H. Zhang, J. Barber, et al.
ICML 880 cites
2022
Improving few-shot classification with unlabeled examples
P. Bateni*, J. Barber*, J. Van de Meent, F. Wood
WACV 95 cites
2020
Sparse Gaussian processes via parametric families of compactly-supported kernels
J. Barber
Education
2015
Advanced Study Program
Massachusetts Institute of Technology
Physics, Mathematics, Machine Learning (non-degree program)
2010
M.S. Computer Science
Johns Hopkins University
Cryptography, DSP, Machine Learning, Bayesian Inference
2008
B.S. Computer Science
Penn State University
Technical skills

Languages

PythonRustC / C++TypeScriptJulia

ML Frameworks

JAXPallasPyTorchCUDA

LLM Tooling

Claude SDKCustom Harnesses