Shuang Ma
I am currently a Senior Research Scientist at Apple, specializing in foundational language models. Since joining Apple, my primary focus has been on developing and launching Apple’s large language models (Apple Intelligence Foundation Language Models). My key contributions include designing effective RLHF (Reinforcement Learning with Human Feedback) algorithms, building the RLHF training pipeline, and managing model training for post-training stages. I have also led efforts in data selection and synthetic data generation to support post-training.
Recently, I’ve expanded my focus to improving reasoning capabilities and developing the post-training pipeline for Apple’s multimodal LLMs.
Prior to joining Apple, I was a Senior Researcher at Microsoft Research, where I concentrated on foundational research in multimodal pretraining, representation learning, and building foundation models for embodied agents.
Email: yunyikristy <AT> gmail <DOT> com
Research Interests
-
Foundation models
-
Multimodal Learning
News
-
Our technical report for Apple's LLM is released. 'Apple Intelligence Foundation Language Models'. [Paper]
-
Our paper 'MMAU: A Holistic Benchmark of Agent Capabilities Across Diverse Domains' is released. [Paper][Code]
-
Our paper 'TOOLSANDBOX: A Stateful, Conversational, Interactive Evaluation Benchmark for LLM Tool-Use Capabilities' is released. [Paper][Blog]
-
I am serving as Area Chair for NeurIPS 2024.
-
​Our paper 'TACO: Temporal Latent Action-Driven Contrastive Loss for Visual Reinforcement Learning' is accepted by NeurIPS 2023. [Paper]
-
Our paper 'Is Imitation All You Need? Generalized Decision-Making with Dual-Phase Training' is accepted by ICCV 2023. [Paper][Code]
-
I am organizing workshop "PerDream: PERception, Decision making and REAsoning through Multimodal foundational modeling" at ICCV 2023. [Website]
-
I am co-organizing "Workshop on Robot Learning and SLAM" at ICCV 2023. [Website]
-
I am serving as Area Chair for NeurIPS 2023.
-
Our paper 'PACT: Perception-Action Causal Transformer for Autoregressive Robotics Pre-Training' is accepted by IROS 2023. [Paper][Code]
-
We announced SMART and released our code! [Blog][MSR Research Focus][Code]
-
​​Our paper 'SMART: Self-supervised Multi-task pretrAining with contRol Transformers' is accepted by ICLR 2023 as notable top 25% (Spotlight).
-
Our paper 'LaTTe: Language Trajectory TransformEr' is accepted by ICRA 2023. [Blog][Code]
-
Our team EgoMotion-COMPASS got the 2nd place on two tasks of Ego4D challenge (ECCV 2022).
Interns
-
Jianchen Lei, Zhejiang University 2023
-
Yao Wei, Zhejiang University 2023
-
Ruijie Zheng, University of Maryland 2023
-
Yanchao Sun, University of Maryland 2022
-
Arthur Fender Coelho Bucker, Technical University of Munich (TUM) 2022
-
Weijian Xu, UC San Diego 2021
-
Cherie Ho, Carnegie Mellon University 2021
-
Zhaoyang Zeng, Sun Yat-sen University 2020
-
Mingzhi Yu, University of Pittsburgh 2020