CV

Graduate Researcher

University of Southern California

Supervisor: Dr. Payam Piray

• Extended a popular predictive-representation-based reinforcement learning model (default representation) with a novel, feature-based approach. The extension allows the key element of compositionality to be maintained in a fully feature-based definition, and we mathematically proved convergence (NeurIPS, under review).
• Extended reinforcement learning models for brain decision-making, achieving efficient online learning with policy-independent environmental representations for task transfer. This provides a normative and unified explanation for seemingly disparate behavioral differences in human decision-making tasks (Nature Communications, under review).
• Implemented an episodic memory buffer into a popular RL world model (DreamerV3) and showed that a simple episodic memory component helps in environments that require long-term memory.
• Experimented with different types of deep RL models (model-based and model-free) as encoding models for the brain, comparing the predictive accuracy of each to see whether there was a distinction in the brain.

Research Engineer

Honda Research Institute

Supervisor: Dr. Behzad Dariush

• Developed a unified framework for the joint problem of trajectory prediction and imputation — the pioneering work that identified and filled the gap in benchmarks and techniques for this joint problem (CVPR 2023).
• Developed predictive models for road scene classification and segmentation into an orthographic bird’s-eye view (BEV) frame. Explored temporal architectures to handle long- and short-term context and address occlusions.
• Assisted in deploying a real-time trajectory prediction model onto a vehicle, including model fine-tuning with real-world data, optimizing pre- and post-processing for efficiency, and adapting the inference pipeline for hardware constraints.

Undergraduate Researcher

Center for Artificial Intelligence in Society (CAIS), USC

Supervisor: Dr. Bistra Dilkina

• Assisted in implementing and testing an unsupervised clustering and classification network parameterized by a Gaussian mixture model for camera-trap images. Tested the approach against state-of-the-art clustering benchmarks such as DEC and DeepCluster.

Undergraduate Researcher

Neuro Imaging With Deep Learning Lab (NIDL), USC

Supervisor: Dr. Hosung Kim

• Artificially aged brains on fMRI images using GANs to identify neurodegeneration. Improved U-Net and Capsule Net segmentation by adding multi-scale highlighted foregrounds for white matter hyperintensities.
• Established a standardized image processing pipeline for MRI images using a shell script with FSL commands.

Undergraduate Researcher

Autonomous Networks Research Group (ANRG), USC

Supervisor: Dr. Bhaskar Krishnamachari

• Proposed and implemented a novel ML solution to estimate vehicular communication ranges in dynamic systems.

Software Engineering Intern

Qualcomm

Supervisor: Dr. Suhail Jalil

• Created a C-sim for facial recognition hardware and converted existing code into C++. Produced a parsing script to optimize product testing for current and future engineers.

Software Engineering Intern

Qualcomm

Supervisor: Dr. Suhail Jalil

• Used machine learning on capacitive signals from touch screens to identify whether a user was an adult or a child. The project was accepted to the Qualcomm Machine Learning Summit for Fall 2017.

PhD, Computational and Cognitive Neuroscience

University of Southern California

• Expected 2028.

MS, Electrical Engineering

University of Southern California

• 3.8 GPA.

BS, Electrical Engineering

University of Southern California

Default Feature Representations of the Cognitive Map

Armin Bazarjani, Payam Piray — under review at NeurIPS 2026

Efficient Learning of Predictive Maps for Flexible Planning

Armin Bazarjani, Payam Piray — under review at Nature Communications

Uncovering the Missing Pattern: Unified Framework Towards Trajectory Imputation and Prediction

Yi Xu, Armin Bazarjani, Hyung-gun Chi, Chiho Choi, Yun Fu — Conference on Computer Vision and Pattern Recognition (CVPR)