Max Simchowitz

Timeline

1. By 2015

   Mathematics at Princeton

   By 2015, he majors in mathematics at Princeton University and researches with Sanjeev Arora and David Blei. Princeton awards him the George B. Covington Prize for Excellence in Mathematics in 2015.

2. 2018

   Fairness and delayed impact

   With Lydia T. Liu, Sarah Dean, Esther Rolf and Moritz Hardt, he coauthors “Delayed Impact of Fair Machine Learning,” which wins ICML’s Best Paper Award in 2018. The paper finds that common fairness criteria may fail to improve outcomes over time and can cause harm. Berkeley also awards him its Tong Leong Lim Pre-Doctoral Prize in 2018.

3. May 2021

   PhD on linear control

   He completes a PhD focused on theoretical questions in machine learning in Berkeley’s EECS department in May 2021, co-advised by Ben Recht and Michael Jordan. His dissertation, “Statistical Complexity and Regret in Linear Control,” shows that a simple strategy achieves optimal regret for adaptive control of the linear quadratic regulator.

4. 2022

   An outstanding paper at ICML

   After completing his 2021 PhD, he joins Russ Tedrake’s Robot Locomotion Group at MIT CSAIL as a postdoctoral researcher. With Hyung Ju Suh, Kaiqing Zhang and Tedrake, he coauthors “Do Differentiable Simulators Give Better Policy Gradients?”, an ICML Outstanding Paper in 2022. The paper shows that stiffness or discontinuities can weaken first-order estimates and proposes an alpha-order estimator that combines exact gradients with zero-order methods.

5. 2024

   A Simons Institute fellowship

   He spends fall 2024 as a VMware Research Fellow in the Simons Institute’s Modern Paradigms in Generalization program. In October 2024, he is a former MIT postdoc and incoming assistant professor at Carnegie Mellon.

6. September 2024

   DPPO for robot policy tuning

   In September 2024, he coauthors “Diffusion Policy Policy Optimization” (DPPO) with Allen Z. Ren and seven other collaborators. The framework fine-tunes diffusion-based robot policies with policy-gradient reinforcement learning and demonstrates zero-shot deployment of simulation-trained policies on robot hardware in a long-horizon, multistage manipulation task.

7. October 2024

   Diffusion Forcing for robots and video

   In October 2024, he coauthors “Diffusion Forcing: Next-token Prediction Meets Full-Sequence Diffusion” with Boyuan Chen, Diego Martí Monsó, Yilun Du, Russ Tedrake and Vincent Sitzmann. The method adds a different noise level to each token while predicting later tokens, enabling flexible robot plans and stable video generation.

8. Since 2025

   Professor at Carnegie Mellon

   In 2025, he joins Carnegie Mellon University’s Machine Learning Department as an assistant professor and serves as courtesy faculty in its Robotics Institute. His lab studies how to help AI learn quickly and generalize to unseen scenarios, drawing on control theory, reinforcement learning, deep learning, generative modeling and optimization. He collaborates closely with Maven Robotics, Mimic Robotics and the robotics team at Toyota Research Institute.