I am currently a Research Scientist at Google DeepMind.
I work on post-training for Gemini's coding capabilities,
with a particular focus on SWE agents. Earlier at Google, I worked on Jules.
Previously, I was a Principal Researcher at Microsoft Semantic Machines,
where my work leveraged large language models to enable scalable construction of conversational AI systems.
I also worked on other projects relating to privacy, model compression, and crowdsourcing,
in collaboration with interns and teammates.
I received my PhD in Computer Science
at UC Berkeley,
where I was advised by Dawn Song.
I was a member of the Berkeley AI Research Lab and have also collaborated with the RISE Lab.
I also received my MS and BS degrees at UC Berkeley.
I've worked at
Google AI,
Intel Labs,
and Microsoft Research AI.
In the past, I have also done research relating to security applications of
machine learning, software security, and web security.
Papers
2025
Gemini 2.5: Pushing the frontier with advanced reasoning, multimodality, long context, and next generation agentic capabilitiesGemini Team, Google
arXiv
2024
Learning to Retrieve Iteratively for In-Context LearningEMNLP 2024
Language-to-Code Translation with a Single Labeled ExampleEMNLP 2024
Privacy-Preserving In-Context Learning with Differentially Private Few-Shot GenerationICLR 2024
2023
BenchCLAMP: A Benchmark for Evaluating Language Models on Semantic ParsingNeurIPS 2023, Datasets and Benchmarks Track
ToolTalk: Evaluating Tool Usage in a Conversational SettingarXiv
Privacy-Preserving Domain Adaptation of Semantic ParsersACL 2023
2022
Few-Shot Semantic Parsing with Language Models Trained On CodeNAACL 2022 (short paper)
Guided K-best Selection for Semantic Parsing AnnotationACL 2022 (demo track)
Addressing Resource and Privacy Constraints in Semantic Parsing Through Data AugmentationFindings of ACL 2022
2021
Pruning Pretrained Encoders with a Multitask ObjectiveENLSP workshop at NeurIPS 2021
Constrained Language Models Yield Few-Shot Semantic ParsersEMNLP 2021
2020
RAT-SQL: Relation-Aware Schema Encoding and Linking for Text-to-SQL ParsersACL 2020
2019
Hierarchical Variational Imitation Learning of Control ProgramsarXiv
Program Synthesis and Semantic Parsing with Learned Code IdiomsNeurIPS 2019
Synthetic Datasets for Neural Program SynthesisICLR 2019
2018
Hierarchical Imitation Learning via Variational Inference of Control ProgramsInfer2Control workshop at NeurIPS 2018
Improving Neural Program Synthesis with Inferred Execution TracesNeurIPS 2018 (spotlight presentation)
Imitation Learning of Hierarchical Programs via Variational InferenceNAMPI workshop at ICML 2018 (extended abstract)
Differentiable Neural Network Architecture SearchICLR 2018 workshop track
Towards Specification-Directed Program RepairICLR 2018 workshop track
Parametrized Hierarchical Procedures for Neural ProgrammingICLR 2018
2017
JPEG-resistant Adversarial ImagesMachine Learning and Computer Security workshop at NeurIPS 2017
PIANO: Proximity-based User Authentication on Voice-Powered Internet-of-Things DevicesICDCS 2017
Making Neural Programming Architectures Generalize via RecursionICLR 2017 (best paper award)
2016
ExploreKit: Automatic Feature Generation and SelectionICDM 2016 (short paper)
Latent Attention for If-Then Program SynthesisNeurIPS 2016
2015
Exploring Privacy Preservation in Outsourced K-Nearest Neighbors with Multiple Data OwnersCCSW at CCS 2015
Recognizing Functions in Binaries with Neural NetworksUSENIX Security 2015
2014
Joint Link Prediction and Attribute Inference Using a Social-Attribute NetworkTIST, published 2014-04
2012
On the Feasibility of Internet-Scale Author IdentificationIEEE S&P 2012
FreeMarket: Shopping for free in Android applicationsNDSS 2012 (extended abstract)
2011
A Systematic Analysis of XSS Sanitization in Web Application FrameworksESORICS 2011
2010
Inference and Analysis of Formal Models of Botnet Command and Control ProtocolsCCS 2010
The Emperor's New APIs: On the (In)Secure Usage of New Client-side PrimitivesW2SP workshop at IEEE S&P 2010
