Fred Xu

Ph.D. in Computer Science

University of California, Los Angeles

B.S. in Computer Science and Engineering, minor in Mathematics

University of California, Los Angeles

AI Research Intern

Block, Inc. · Oakland, CA

• Function-space flow matching. Proposed kernel Functional Flow Matching (kFFM), which replaces independent prior–data pairing with entropic optimal transport under a kernel-induced cost, with well-posedness and discretization-invariance guarantees; improved distributional matching over FFM, diffusion and OT baselines on time-series and PDE benchmarks. First-author paper at NeurIPS 2026.
• Graph anomaly detection for fraud. Developed EB-GAD, a training-free, label-free anomaly detector that fits a graph prior by empirical Bayes and scores nodes by closed-form finite-horizon control energy; best or tied-best AUROC on 9 of 11 benchmarks, including four financial fraud networks of up to 3.7M nodes, with margins of up to 21.7 points. First-author paper at NeurIPS 2026.
• Uncertainty quantification for GNNs. Proposed DSS-GNN, which expands node embeddings on graph-spectral and Wiener-chaos axes so a single representation supports calibrated prediction, out-of-distribution detection and distribution-shift robustness. First-author paper at NeurIPS 2026.
• Graph generation. Co-developed FlowPhon, which generates dense graphs via flow matching on graphon operators; second-author paper under review.
• Transaction foundation model. Built a GPT-style foundation model over customer transaction event sequences to model Cash App balance changes.

STEM Specialist

Handshake AI, Inc. · Remote

• LLM post-training data. Authored and solved advanced STEM problems and evaluated LLM-generated code to support alignment and improve the generation quality of LLM agents.

AI Research Intern

Block, Inc. · Oakland, CA

• Uncertainty estimation on graphs. Developed a graph neural network grounded in stochastic partial differential equations for uncertainty estimation on graph data. First-author paper at NeurIPS 2025.
• Customer risk modeling. Implemented a probabilistic machine learning model with Monte Carlo simulation to estimate customer default risk and account lifetime value.
• Time-series generation. Created a distributional diffusion model for time series built on path signatures.

Research Scientist Intern

Hitachi America, Ltd. · Santa Clara, CA

• Sensorless cargo monitoring. Engineered features and built regression and time-series models on global cargo-container temperature and humidity data to estimate shipment operational risk, improving on the baseline model by 30%.
• Financial risk simulation. Ran Monte Carlo simulations over financial indicators and corporate portfolios, combined with time-series analysis, to estimate corporate financial risk under currency-rate and market shifts.

Applied Scientist Intern

Amazon.com, Inc. · New York, NY

• Audience embedding for recommendation. Implemented a deep-learning bipartite graph embedding model for audience representation in recommendation and user-group behavior prediction; proposed fuzzy-set embeddings to support logical queries, outperforming the baseline by at least 30%.

Software Engineer

NortonLifeLock Inc. · Culver City, CA

• Privacy Monitor (iOS). Developed the privacy monitor feature and client kit in Swift for the Norton Secure VPN iOS app, increasing customer interest in the privacy product by 45%.
• Norton Secure VPN (Android). Maintained and extended the Android app (10M+ installs), shipping features including split tunneling and torrenting support.
• Norton Security UWP. Developed, integrated and unit-tested core C#, Angular and C++/WinRT code for the Universal Windows Platform app; helped build the product threat model and implemented encryption, authentication and file-integrity mechanisms.
• Secure file transfer and privacy MVPs. Built a cross-platform MVP (Android, iOS, PWA) for secure file transfer using TLS, RSA, AES-256 and SHA-256 with key management in Azure Key Vault, and a Chrome extension that opts users out of data-broker websites.

Instructor, CS 97: Introduction to Data Science

UCLA Summer Institute

Taught advanced high-school students for college credit; rated 8.75 / 9 overall.

Teaching Assistant

UCLA Computer Science

CS 148 Introduction to Data Science, CS 247 Advanced Data Mining, CS 33 Computer Organization, CS M51A Logic Design of Digital Systems; average rating 8.4 / 9.