Sriram Somasundaram
About
Detail
Co-Founder
San Francisco, California, United States
Résumé
Jobs
verified_user
0% verified
Co-FounderLatentOct 2022 - Current (3 years 9 months)Backed by Y Combinator and working with several of the largest US health systems. We're hiring!- Founding EngineerHebbiaJan 2021 - Oct 2022 (1 year 10 months)Backed by Index, Floodgate, Peter Thiel, founders of Yahoo, and first investors in Google.
- Graduate Teaching AssistantStanford UniversityJan 2019 - Dec 2021 (3 years)TA for CS236 Deep Generative Models (Prof. S. Ermon, A. Grover, Fall 2019) TA for CS109 Probability for Computer Scientists (Prof. D. Varodayan, Winter 2020 and Prof. L. Yan, Spring 2020)
- Graduate Research StudentStanford Artificial Intelligence Laboratory SAILJan 2019 - Dec 2021 (3 years)Stanford Vision and Learning Lab - People, AI, and Robots Group, projects advised by Prof. Fei-Fei Li. Cutting edge research in deep reinforcement learning, robotics, meta-learning, and computer vision. Task-specific exploration via uncertainty-driven curiosity during inference time. Interactive physical and semantic understanding for multi-step robotic manipulation with an object-centric latent state estimation architecture and model-free RL baselines.
- Software EngineerRiot GamesMay 2018 - Aug 2019 (1 year 4 months)Esports - we aspire to bring joy to billions of esports fans around the world. Our mission is to transform and enrich digital experiences through crafting products and experiences that seek to achieve maximum player value
- UResearcherUSC Cognitive Learning for Vision and Robotics CLVRJan 2018 - Feb 2019 (1 year 2 months)Neural Program Synthesis from Diverse Demonstration Videos (ICML 2018) - Built a neural program synthesizer that programmatically describes behavior in demo videos. Trained with a multi-task objective to induce meaningful latent representations and a two-path LSTM to summarize and capture diverging conditions in video. Composing Complex Skills by Learning Transition Policies (ICLR 2019) - Enabled smooth skill composition in hierarchical reinforcement learning by transitioning between learned skills. Trained a discriminator as a proxy for a reward model and transitioned b/w complex skills without dense signals.
- ResearcherStanford University School of MedicineJan 2013 - Dec 2015 (3 years)Investigated antigen presentation with pH dependent DM/DO involving T2 & S2 cell cultures, protein purification, HPLC, column chromatography, western blot, coomassie, ELISA, and more. Worked on a narcolepsy project involving sequencing single cells from narcolepsy patients through single cell sorting and barcoded PCR. Passion project to synthesize chitosan nanoparticles for an ocular drug delivery system. Did microscopy, drug loading and release studies, permeation using franz cell and collagen gel, protein docking using Autodock vina software, and studies of electric field changes as a result of protonation.
Education
verified_user
0% verified
- BBachelor of Science - BS, Computer Science
- High SchoolThe Harker School
- Master of Science - MS, Computer Science (Artificial Intelligence)Stanford UniversityWas selected as one of the strongest incoming Masters students and offered full financial support by the CS department
Projects (professional or personal)
verified_user
0% verified
- LLearning CarAug 2015 - Current (10 years 11 months)Implemented reinforcement learning (RL) in a robotic car. Used the goBetwino API to connect an Arduino Mega microcontroller to RL program which received car state info from the serial port.
Awards
verified_user
0% verified
- SSumma Cum LaudeDec 2018
- AAdditional HonorsPresidential Scholarship (USC) National AP Scholar - Computer Science, Calculus BC, Chemistry, Spanish, World History, U.S. History, Biology, Microeconomics, Macroeconomics, Psychology, Environmental Science, Statistics, Physics C: Mechanics, Physics C: Electricity and Magnetism National Merit Finalist 2015 Intel STS Semifinalist Intel International Science and Engineering Fair (ISEF) - 2nd place (2014) in Biochemistry First place at Synopsys Championship Fair (2014) & Grand Prize – Best of Biological Sciences n+1 award at Synopsys
Publications
verified_user
0% verified
- IComposing Complex Skills by Learning Transition Policies with Proximity Reward InductionInternational Conference on Learning Representations ICLRJan 2019Intelligent creatures acquire complex skills by exploiting previously learned skills and learning to transition between them. To empower machines with this ability, we propose transition policies which effectively connect primitive skills to perform sequential tasks without handcrafted rewards. To effectively train our transition policies, we introduce proximity predictors which induce rewards gauging proximity to suitable initial states for the next skill. The proposed method is evaluated on a diverse set of experiments for continuous control in both bi-pedal locomotion and robotic arm manipulation tasks in MuJoCo. We demonstrate that transition policies enable us to effectively learn complex tasks and the induced proximity reward computed
- INeural Program Synthesis from Diverse Demonstration VideosInternational Conference on Machine Learning ICMLJan 2018Interpreting decision making logic in demonstration videos is key to collaborating with and mimicking humans. To empower machines with this ability, we propose a neural program synthesizer that is able to explicitly synthesize underlying programs from behaviorally diverse and visually complicated demonstration videos. We introduce a summarizer module as part of our model to improve the network’s ability to integrate multiple demonstrations varying in behavior. We also employ a multi-task objective to encourage the model to learn meaningful intermediate representations for end-to-end training. We show that our model is able to reliably synthesize underlying programs as well as capture diverse behaviors exhibited in demonstrations. Website an
- SpH-susceptibility of HLA-DO tunes DO/DM ratios to regulate HLA-DM catalytic activityScientific Reports NatureNov 2015