About Me
Hi, my name is Sima and I’m a Senior Staff AI Scientist at Samsung SDS AI Science Lab in San Jose CA. Prior to Samsung, I worked as a Data Scientist, and Probabilistic Engineer for Applied Materials and GE Global Research. My background is in Bayesian Machine Learning, Deep Learning, and Generative AI. I received my PhD from Georgia Institute of Technology on Topology Optimization for Computer-Aided Engineering (CAE) of microstructures under the supervision of Professor Tequila Harris.
Experience
AI Science Lab, San Jose, CA, USA
-
Responsibilities : Establishing technical goals, Leading task force & collaborations across Samsung’s or external entities for Technology Development & Technology Transfer, Developing products, solutions & services to secure Samsung’s competitive advantages.
-
Projects:
- Generative AI:
- Development of tailored multi-modal LLMs for data in text, image, and tabular formats
- Uncertainty quantification and model calibration for text-image multi-modal and large language models
- Data efficient Deep Learning and AI:
- Development of Representation & Unsupervised Learning algorithms for reducing the labeling need for Deep Learning models
- Development of data efficient alignment methods for LLMs
- Computer Vision:
- Active Learning and Unsupervised accuracy estimation for classification, detection & semantic segmentation
- Probabilistic Multi-View 3D semantic segmentation Deep Learning Active Learning framework
- Predictive data analytics:
- Developed sale forecasting models for Samsung business units
- Optimized marketing strategies for sale campaigns
- Generative AI:
Data Science Group, Santa Clara, CA, USA
- Development & deployment of AI solutions to increase the performance/ productivity of engineering, services & supply chain teams
- Leading AI app productization in collaboration with Software, UI, DevOp teams & internal or external customers
Probabilistic Lab, San Ramon, CA, USA
- Development of Probabilistic Methods & Machine Learning models for calibration & validation, uncertainty quantification, optimization and meta-modeling/surrogates modeling
- Development of Deep Learning & Machine Learning models to optimize engineering design & manufacturing process
- Development of Computer Vision defect detection software
Education
Georgia Institute of Technology
Ph.D. & M.S., Mechanical Engineering
2009 - 2014
Research topics: CAE, Topological Characterization & Optmization, Image Proessing
- Developed periodic surface parametric models to represent the topology of microstructures
- Developed voxel-based representations for 3D composite microstructures
- Conducted topological characterization of 3D models using graph representations
- Conducted topology optimization for voxel representation of 3D volumes using GA
University of Tehran
B.S., Mechanical Engineering
2001 - 2005
Selected Publications
- My full publication list can be found at Scholar profile.
Here is the list of my recent publictions:
- Improving instruction following in language models through proxy-based uncertainty estimation, arXiv, Accepted at ICML, (2024)
- Bayesian active learning for semantic segmentation, arXiv, (2024)
- Self-Supervised contrastive representation learning for 3D mesh segmentation, arXiv, Accepted at AAAI, (2023)
- Highly efficient representation and active learning framework for imbalanced data and its application to COVID-19 X-Ray classification, arXiv NeurIPS Data-Centric AI workshop, (2021)
- Active learning performance in labeling radiology images is 90% Effective, Frontiers in Radiology, (2021)
- PatchNet: Unsupervised object discovery based on patch embedding, arXiv, (2021)
- Modeling and optimizing the impact of process and equipment parameters in sputtering deposition systems using a Gaussian process machine learning framework, IEEE Transactions on Semiconductor Manufacturing, (2021)
Selected Patents
- Bayesian semantic segmentation active learning with Beta approximation, 20230368507
- Object discovery, 20220383105
- Unsupervised representation learning and active learning to improve data efficiency, 20220138935
- Long short-term memory anomaly detection for multi-sensor equipment monitoring, 20200104639
- Correcting component failures in ion implant semiconductor manufacturing tool, 11348813
- Chamber matching with neural networks in semiconductor equipment tools, 11133204
- Magnetic mixer, 11097236
- Magnetic drive for bioreactor, 10335750
- System and method for characterizing conditions in a fluid mixing devices, 10682618
Awards & Recognitions
- Samsung SDS Circle of Excellence Award granted by CEO of Samsung SDSA for Outstanding performance, Creativity, Organizational abilities and Team Work, (2022)
- Best Presentation Award at Future of Information and Communication Conference (FICC), for the paper titled “Medical Image Labeling via Active Learning is 90% Effective”, (2022)
- GE Above & Beyond Award for development of automated probabilistic testing framework, (2017)
- GE Global Research Aero-Thermal Mechanical System (ATMS) impact award for outstanding contribution to Artificial Lift Optimization project, (2016)
- ASME’s International Engineering Congress and Exposition outstanding research award, (2012)
Projects
Proxy-Based Uncertainty Estimation for Improving Instruction Following in Language Models
arXiv linkAccepted at ICML 2024
We defined a novel Uncertainty-aware Reward Model (URM) for the preference training of LLMs based on Bayesian approximation to quantify the uncertainty of paired responses. We experimentally demonstrated that using URM in LLMs training boosts their instruction following capability and their policy optimization objectives. The URM based finetuning surpasses existing methods by a large margin on benchmarks such as Vicuna and MT-bench.
- The code & data for the paper are shared at P-B-U Git Repo
