Mythic, Sr. Software Performance Optimization Engineer, AI Engineering Team (June 2021 – present)

Apple (Contractor), AI/ML Education Moderator, AI Education Team (January 2020 – April 2021)

• Led the development of a containerized TensorFlow.JS-based image classification web app using Docker and Kubernetes.
• Supported the development of Apple's first deep learning course in a cross-functional team collaborating with engineers from the Watch, Manufacturing, and Battery Optimization teams.
• Designed and trained several XGBoost and TensorFlow/PyTorch models to tackle real-world business problems including manufacturing defective product image classifiers and user-behavior characterization algorithms to extend battery life.
• Created new python-based coding assignments using TuriCreate, TensorFlow, and PyTorch in various topics including sentiment analysis, recommender systems, outlier detection, and transfer learning to enhance existing coursework.
• Generated coding utilities and tutorials for internal infrastructure tools including S3 blob storage and big data management systems, cloud computing for distributed training, and optimization/quantization methods for on-device deployment.
• Mentored and helped learners in various internal courses offered through the AI Education organization with subjects in ML foundations, regression, classification, and clustering and information retrieval.

University of California, Berkeley (August 2017 – May 2021)

Graduate Student Researcher, Neutronics Team (August 2018 – May 2021)

• Created, released, and documented NucML, the first end-to-end python supervised machine learning pipeline for enhanced bias-free nuclear data generation and evaluation to support the advancement of innovative nuclear systems.
• Developed and deployed EXFOR SQL, a modernized version of the EXFOR database on Google BigQuery for accelerated analysis. Data analysis rates are 2x-10x faster than traditional methods using NucML’s automated workflow pipeline.
• Designed, trained, and optimized several algorithms including GBM and Neural Networks for inference of low- and high-energy nuclear reaction probabilities using Scikit-Learn, XGBoost, and TensorFlow. Using ML-generated reaction probabilities, benchmark calculations resulted in an accuracy boost of up to 170% compared to established methods.
• Inferences using trained KNN/DT/XGB models for new data measurements outperformed theoretical models by 170%.
• Provide alternative model selection frameworks for experimental-based datasets that has enhance generalization by 25%.

Graduate Student Researcher, Complexity Team (January 2018 – May 2019)

• Designed and carried out data acquisition campaigns at UC Berkeley’s Cyclotron using sensor kits capable of capturing temperature, acceleration, magnetic field, pressure, humidity, and proximity time-series data to detect operational status.
• Using feature importances, temperatures were found to be correlated with operation times. By querying and collecting local weather data, false correlations due to the time-of-day were eliminated to create a real-detection scenario.
• A DNN classifier architecture was trained resulting in phi coefficients greater than 0.90 for interior-based data. The use of exterior sensor data caused the phi coefficient to drop to 0.72 due to the importance of proximity infrared-based data.

Graduate Student Researcher, Nuclear Engineering Department (August 2017 – May 2018)

• Created automatic data collection pipelines to extract California’s electricity demand and consumption time series data.
• Based on past trends on the collected data, energy, heat, and hydrogen generation dynamics on an in-house designed Nuclear-Renewable Hybrid Energy System (NR-HES) were modeled and optimized for cost using MATLAB Simulink.
• Hydrogen production costs were reduced to $1.66/kg, 32% below the Department of Energy target. The low production costs coupled with potential profits resulted in a 40% reduction in the overall levelized cost of electricity ($0.0199/kWh).

Oak Ridge National Laboratory, ASTRO Researcher, RNSD Division (June 2019 – September 2019)

• Developed a python-based automatic analysis pipeline (SCALE PyTools) capable of performing parametric burnup and criticality calculations using SCALE and SERPENT, external multi-physics tools for simulating nuclear reactor systems.
• The new functionalities in SCALE for addition and removal of material were tested and validated using experimental data from the Molten Salt Reactor Experiment.
• Results obtained showed a deviation of 6.9% from the experimentally calculated poison fraction making it the first simulation benchmark of its kind. Discovered best practices were released publicly as a technical report.

Tokyo Institute of Technology, Graduate Student Intern (Summer 2018)

• Worked on the analysis of the Japan Sodium-Cooled Fast Reactor for minor actinide transmutation with high proliferation resistance.
• Reactivity and burnup calculations for core analysis were completed using CITATION, a 2-D deterministic code. SLAROM was also used to generate the average multi-group cross sections (cell homogenization).
• Parametrically studied decay heat and spontaneous fission isotopic barriers for enhanced proliferation resistance of core blankets.

University of California, Berkeley (Expected: May 2021)
PhD Candidate in Nuclear Engineering (Gester Global Solutions Fellowship)
Thesis: Machine Learning-augmented Nuclear Data Evaluation

University of California, Berkeley
Master of Engineering in Nuclear Engineering (CONACYT Full Scholarship)

Universidad de las Américas, Puebla
Bachelor of Science in Nanotechnology and Molecular Engineering (Honors Graduate)

• Vicente-Valdez, P., Bernstein, L., & Fratoni, M. (2021). Nuclear Data Evaluation Augmented by Machine Learning. Annals of Nuclear Energy. (Submitted)
• Vicente-Valdez, P., Bernstein, L., & Fratoni, M. (2021). NucML: Python Package for ML-based Nuclear Data Cross Section Evaluations. ANS Annual Meeting. (Submitted and Accepted)
• Vicente-Valdez, P., Bernstein, L., & Fratoni, M. (2021). U-233 Nuclear Data Evaluation using Machine Learning Generated Cross Sections. ANS International Conference on Mathematics and Computational Methods Applied to Nuclear Science and Engineering. (Submitted and Accepted).
• Vicente-Valdez, P., Bernstein, L., & Fratoni, M. (2020). Application of Machine Learning to Nuclear Data Evaluation. ANS Virtual Winter Meeting, 123, 1287–1290. https://doi.org/10.13182/T123-32998
• Vicente-Valdez, P., Betzler, Benjamin R., Wieselquist, William, & Fratoni, Massimilliano. Modeling Molten Salt Reactor Fission Product Removal with SCALE. United States. https://doi.org/10.2172/1608211
• Stewart, C., Goldblum, B. L., Chockkalingam, S., Padhy, S., Tsai, Y. A., & Valdez, P. V. & Wright, A. (2019). Multimodal Data Analytics for Nuclear Facility Monitoring. INMM Annual Meeting. Institute of Nuclear Materials Management.

• Machine Learning with TensorFlow on Google Cloud Platform Specialization – Google Certificate
• IBM Applied AI Professional Certificate – IBM Certificate
• IBM Data Science Professional Certificate – IBM Certificate
• Deep Learning Specialization – deeplearning.ai Certificate
• DeepLearning.AI TensorFlow Developer – deeplearning.ai Certificate
• TensorFlow: Data and Deployment Specialization – deeplearning.ai (in progress)
• TensorFlow: Advanced Techniques Specialization – deeplearning.ai (in progress)
• Accelerated Computer Science Fundamentals with C++ Specialization – University of Illinois (in progress)
• Modern Big Data Analysis with SQL Specialization – Cloudera (in progress)

• Gester Global Solutions Fellowship (2020)
• JASSO Scholarship, Japan Student Services Organization (2018)
• Fung Institute Grant, University of California at Berkeley (2017)
• CONACYT-Sener Full Scholarship Award, Science and Technology National Council (2017)
• Honors Program Research Award, Universidad de las Americas Puebla (2013)
• Academic Excellence Scholarship, Universidad de las Americas Puebla (2012)

Fluent in Spanish and English, learning French