Ava Kouhana

Ava Kouhana

MS Student — Stanford ICME · Computational & Mathematical Engineering

I am an ICME master's student at Stanford University. Prior to Stanford, I spent six months at Harvard under Dr. Mengyu Wang working on image segmentation and vision-language models for medical imaging fairness, and six months with Dr. Craig Levin applying latent diffusion models to PET image reconstruction. My research interests span computer vision, deep learning, generative AI, and robotics.

Publications

FairSeg
ICLR 2024

FairSeg: A Large-Scale Medical Image Segmentation Dataset for Fairness Learning Using Segment Anything Model with Fair Error-Bound Scaling

Yu Tian*, Min Shi*, Yan Luo*, Ava Kouhana, Tobias Elze, Mengyu Wang

Paper Code Dataset
FairCLIP
CVPR 2024

FairCLIP: Harnessing Fairness in Vision-Language Learning

Yan Luo*, Min Shi*, Muhammad Osama Khan*, Muhammad Muneeb Afzal, Hao Huang, Shuaihang Yuan, Yu Tian, Luo Song, Ava Kouhana, Tobias Elze, Yi Fang, Mengyu Wang

Paper Code Dataset
PET Attenuation
IEEE NSS MIC 2024

Direct Generation of Attenuation and Scatter Correction of Brain PET Data Using a Conditional Latent Diffusion Model

Ava Kouhana, M. Jafaritadi, G. Chinn, C.S. Levin

Abstract Poster
Super-Resolution
IEEE NSS MIC 2024

Super-Resolution Tomographic Image Reconstruction Using Latent Diffusion Models

Ava Kouhana, M. Jafaritadi, G. Chinn, C.S. Levin

Abstract Poster

Projects

Residual RL
Robotics · RL

Residual RL with Uncertainty for Dexterous Manipulation

Analyzed interaction between imitation learning and residual RL, showing that increased data can degrade performance.

Report
Language Model
NLP · PyTorch

Language Model Components from Scratch (MiniLM, Transformer, BPE)

Personal Project

Implemented Transformer architecture, byte pair encoding, and a minimal language model from scratch in PyTorch.

Transformer BPE
PiDMASound
Systems · Bare-Metal

PiDMASound — Bare-Metal Audio Engine (Raspberry Pi)

Personal Project

Built a real-time audio system using DMA, PWM, interrupts, and a double-buffered DMA pipeline to generate analog sound from a single GPIO pin.

GitHub

Grants & Awards