Research Assistant Machine Learning

Amir Basareh

Machine Learning Researcher & Computational Scientist

********@******.*** (765) 767–1930 West Lafayette, IN linkedin.com/in/amirhossein-basareh github.com/magronox Summary

Doctoral candidate investigating reasoning capabilities in AI systems, spanning neural and proba- bilistic approaches. Current research focuses on transformer memorization as a fundamental capabil- ity, MCMC-based discrete reasoning, and automated scientific knowledge extraction. Experienced in training diagnostics, ablation studies, and large-scale experiments with PyTorch. Work combines empirical investigation of model capabilities with building reasoning systems. Education

Purdue University, Ph.D. in Computer Science (GPA: 3.66/4.00) Jan 2022–Dec 2026 Sharif University of Technology, B.Sc. in Electrical Engineering (GPA: 3.9/4.0) Aug 2016– May 2021

Technical Skills

• Machine Learning: training-curve diagnostics, hyperparameter tuning, ablation studies, data preprocessing, metrics debugging, experiment tracking (Weights&Biases).

• Frameworks/Tools: PyTorch, JAX, TensorFlow, TVM, NumPy, Pandas, scikit-learn, Git, SLURM.

• Methods: transformers/LLMs, probabilistic modeling & MCMC, algorithmic optimization, graph and tensor computation, HPC techniques.

Experience & Projects

Graduate Research Assistant Purdue University

Scientific Knowledge Extraction & Validation Oct 2025–Present

• Developing an automated system combining LLMs and symbolic modules for extracting and validating scientific knowledge in science and technology domains. Graduate Research Assistant Purdue University

Enigma: Privacy-Preserving QAOA Aug 2025

• Co-authored a paper on privacy-preserving quantum approximate optimization algorithms; led the design of attack models (ML/graph inference) to test obfuscation robustness.

• Submitted to ASPLOS; currently under peer review. Collaboration with CU Boulder. Graduate Research Assistant Purdue University

Transformer Memorization Capacity & Scaling Laws May 2025–Present

• Discovered power-law relationships between sample complexity and model dimension in trans- formers; conducted large-scale GPU experiments with PyTorch/CUDA; preprint ready and awaiting next conference deadline.

• Implemented reproducible training pipelines and tracked experiments with Weights&Biases. 1

Graduate Research Assistant Purdue University

SABLE: Sparse Matrix Computations Oct 2024–Present

• Engineered a dense-region partitioner and integrated it into SABLE’s staging of sparse matrix- vector multiplication.

• Collaborated with faculty advisers to optimize memory traffic and branch divergence; paper rejected once, in revision for next submission.

Graduate Research Assistant Purdue University

Program Synthesis for ARC via MCMC Oct 2024–Present

• Designed a novel MCMC solver for the ARC benchmark focusing on discrete and visual reasoning; uses a directed acyclic graph reasoning framework.

• Developed proposal mechanisms that improved synthesis accuracy and search efficiency. Data Science Intern ProcessMiner Inc., Atlanta, GA May 2024–Aug 2024

• Designed and deployed attention-based, LSTM, and CNN models for multivariate time-series forecasting, improving accuracy by approximately 30%.

• Integrated XGBoost/ARIMA models and developed anomaly detection pipelines.

• Diagnosed production issues and collaborated on code reviews. Graduate Research Assistant Purdue University

UpDown: Hardware-Aware Graph Algorithms Dec 2022–Apr 2024

• Optimized PageRank, BFS, graph convolution and triangle counting algorithms for heteroge- neous architectures through memory-hierarchy-aware dataflows. Graduate Teaching Assistant Purdue University

Jan 2022–Present

• Assisted with CS240 (Programming in C), CS381 (Algorithms), and CS571 (Artificial Intelli- gence); led recitations and provided student support. Graduate Research Assistant Purdue University

Matrix Computations & Tensor Optimization Jan 2022–Aug 2022

• Investigated spectral graph theory and tensor decompositions for large-scale sparse graphs; de- veloped iterative solvers and higher-order methods. Research Assistant Sharif University of Technology Meta-Learning Theory Sep 2020–Aug 2021

• Analyzed the generalization behaviour of Reptile and MAML across few-shot tasks.

• Studied stability and transferability in meta-learned model updates. Publications & Preprints

• Scaling Laws for Transformer Memorization Capacity — preprint ready; awaiting next submis- sion deadline.

• Program Synthesis for ARC via MCMC — manuscript in preparation.

• SABLE: Staging Blocked Evaluation of Sparse Matrix Computations — co-author, arXiv preprint 2407.00829.

• Enigma: Privacy-Preserving Schemes for QAOA — co-author; submitted to ASPLOS. 2

Selected Presentations

• Discrete Reasoning in Transformers; Abstraction and AGI — Purdue Graduate Research Symposium, Spring 2025.

• LLM Reasoning and Thinking Skills — internal group meetings, 2024–2025. Honors & Awards

• National Runner-Up, Data 4Good Competition (Purdue University, 2024).

• Intercultural Diversity & Inclusion Certificate (Purdue University, 2023).

• Silver Medal, National Physics Olympiad (Iran, 2015). 3

Contact this candidate