Aiden Ament

LinkedIn: aidenament
Github: aidenament

Summary

B.S. in Mathematics of Computation, UCLA
Experienced mathematics and machine learning researcher with a focus on deep learning
Proficient in prompt engineering, model evaluation, and performance optimization for large language models
Expertise in applying machine learning to diverse domains including computer vision, genomics, and natural language processing
Skilled in implementing and fine-tuning various neural network architectures including residual networks, recurrent neural networks, and transformers

Use PyTorch to develop novel neural network architectures for computer vision applications
Use compile optimizations and Nvidia Nsight to optimize performance on hardware
Developed model with superior accuracy on imagenet1k (80%) compared to ResNet152 with 20% of the parameters and half the training time

Conducted research on prompt modeling using PHI2, a state-of-the-art small Large Language Model, to evaluate the fairness and factualness of claims on the UniLC dataset
Increased model accuracy from 0.69 to 0.74 with novel prompting techniques
Implemented and tested various evaluation techniques including zero-shot, few-shot, and evidence generation methods to improve model performance
Utilized evidence generation from PHI2 as well as Mixtral 8x7B to enhance PHI2's claim assessment capabilities

Conducted research with deep learning models to predict which DNA sequences proteins will bind to
Leveraged contemporary ML techniques such as using modified residual networks, Bayesian hyper-parameter tuning through Weights and Biases, and slight weight decay in order to classify DNA sequences with near state of the art accuracy
Trained the model on vast genomic datasets, enabling the capture of highly intricate patterns
Implemented advanced data augmentation to increase the amount of available data and improve model generalization

Collaborated with a team of 5 researchers to develop a model estimating entry and exit rates of the homeless population based on observational data collected by civil workers
Coded the model in C++ and implemented multithreading, enabling simulations spanning over a year
Performed statistical analysis of the model, utilizing combinatorics and Bayesian probability estimates to corroborate simulation likelihoods

Spearheaded the establishment of UCLA's first student-led organization dedicated to the responsible development of AI and the mitigation of potential risks from advanced AI systems
Registered the club as a 501(c)(3) nonprofit corporation and raised over $15,000 in funding from Open Philanthropy to support the clubs educational and research initiatives
Organized and led introductory fellowships to teach students the basics of deep learning for computer vision using hands on coding exercises
Directed a research group in mechanistic anomaly detection to improve our understanding of how models come to their conclusions

US Rowing: Competed nationally and internationally throughout high school (2016-2020) and on the UCLA Men's Rowing Team (2020 - 2024)
Studied complex analysis under Terence Tao