Michael Chang

Hi! I’m Michael, a machine learning engineer with over three years of experience turning cutting-edge AI ideas into reality. My passion lies in pushing the boundaries of AI through research in computer vision, reinforcement learning, large language models (LLMs), and multimodal systems.

My goal is to help shape the future of AI as a research scientist at one of the world’s leading research labs, where innovation and curiosity drive progress. Alongside technical work, I thrive in collaborative, fast-paced environments and constantly seek new challenges that expand the boundaries of what AI can achieve.

Beyond research and coding, I recharge by playing volleyball, reading, and exploring the strategic depths of Go.

Publications

A Perspective on Fluid Mechanical Environment for Challenges in Reinforcement Learning

Shruti Mishra, Michael Chang, Vamsi Spandan, Shmuel Rubinstein

RLC 2025 Workshop

ABS PDF CODE

INCLUDE: Evaluating Multilingual Language Understanding with Regional Knowledge

Angelika Romanou, Negar Foroutan, Anna Sotnikova, Zeming Chen, Sree Harsha Nelaturu, Shivalika Singh, Rishabh Maheshwary, Micol Altomare, Mohamed A. Haggag, Snegha A, Alfonso Amayuelas, Azril Hafizi Amirudin, Viraat Aryabumi, Danylo Boiko, Michael Chang, Jenny Chim, Gal Cohen, Aditya Kumar Dalmia, Abraham Diress, Sharad Duwal, Daniil Dzenhaliou, Daniel Fernando Erazo Florez, Fabian Farestam, Joseph Marvin Imperial, Shayekh Bin Islam, Perttu Isotalo, Maral Jabbarishiviari, Börje F. Karlsson, Eldar Khalilov, Christopher Klamm, Fajri Koto, Dominik Krzemiński, Gabriel Adriano de Melo, Syrielle Montariol, Yiyang Nan, Joel Niklaus, Jekaterina Novikova, Johan Samir Obando Ceron, Debjit Paul, Esther Ploeger, Jebish Purbey, Swati Rajwal, Selvan Sunitha Ravi, Sara Rydell, Roshan Santhosh, Drishti Sharma, Marjana Prifti Skenduli, Arshia Soltani Moakhar, Bardia Soltani Moakhar, Ran Tamir, Ayush Kumar Tarun, Azmine Toushik Wasi, Thenuka Ovin Weerasinghe, Serhan Yilmaz, Mike Zhang, Imanol Schlag, Marzieh Fadaee, Sara Hooker, Antoine Bosselut

ICLR 2025

ABS PDF CODE

Projects

Tiny Aya Vision

Part of Expedition Tiny Aya, Tiny Aya Vision is an open-weight multilingual vision-language model under 4B parameters, extending the 3.35B Tiny Aya model (70+ languages) with lightweight visual capabilities through parameter-efficient fusion. It combines a frozen SigLIP2 vision encoder with a small pixel-shuffle + SwiGLU MLP connector and LoRA adaptation of the language model, then merges the LoRA weights back to restore multilingual text performance, following the recipe from Aya Vision. The core question the project investigates: can a small (~3B) multilingual language model gain effective visual grounding through parameter-efficient fusion without sacrificing multilingual text performance or on-device deployability, filling a gap where no existing sub-4B model pairs vision with support for 70+ languages, especially low-resource ones.

CODE BLOG

Open-Rig3R

A reimplementation of Rig3R: Rig-Aware Conditioning for Learned 3D Reconstruction (Li et al., 2025), built for clarity, reproducibility, and extensibility. Rig3R is a transformer-based model for multiview 3D reconstruction and camera pose estimation that incorporates rig-aware conditioning: unlike prior models such as DUSt3R and Fast3R, it learns to leverage rig metadata (camera ID, timestamp, and rig pose) when available, and can infer rig structure when it isn't, enabling robust 3D reconstruction across both unstructured and rig-based image sets. Using a ViT-Large encoder-decoder, the model jointly predicts dense pointmaps, pose raymaps, and rig-relative raymaps, recovers camera poses in closed form from those raymaps, and can perform unsupervised rig discovery, clustering frames by camera and recovering rig extrinsics from unordered image collections.

CODE PAPER