About
Adaptive and efficient AI
I'm a third-year Ph.D. student in Computer Science and Engineering at the University of Notre Dame. I study how AI systems can keep learning as data changes, within practical memory and computation budgets.
My research connects continual learning and efficient inference, from learning algorithms and compact representations to hardware-aware implementation.
Advised by Siddharth Joshi in the Intelligent Microsystems Lab.
Previously, I worked at TCS Research on neuromorphic computing, online learning, and deployment on specialized hardware.
Research
All publicationsEfficient continual learning
Retaining knowledge. Preserving the capacity to learn.
My current research asks how fixed-capacity models can retain useful knowledge and remain able to learn over long sequences of changing tasks. I'm particularly interested in continual learning with bounded auxiliary memory, the balance between stability and plasticity, and learning rules suited to resource-constrained hardware.
Earlier related work
Online time-series forecasting using spiking reservoir
A spiking reservoir with feedback-driven online learning forecasts non-stationary time series in memory-constrained settings.
Efficient inference
Compression that translates into efficient execution.
My work connects low-bit quantization and lossless compression with the kernels and architectures that execute them. The goal is to turn smaller representations into lower memory traffic and faster inference, while accounting for accuracy.
QTEA: Ternary LLMs with Sparse Residual Salient Weight and By-Column Optimization
Ternary weights, structured residual corrections, and efficient kernels connect compact LLM representations with faster inference.
Huff-LLM: End-to-End Lossless Compression for Efficient LLM Inference
Keeping LLM weights compressed across the memory hierarchy reduces storage and bandwidth requirements without changing weight values.
Hardware–software co-design
Designing with memory, precision, and computation in mind.
I study how numerical precision, memory organization, and hardware non-idealities shape the design of efficient AI systems. My work connects model-level evaluation with mixed-signal computation, compute-in-memory architectures, and neuromorphic applications.
Foundry-SRAM-Compatible Compute-Near-Memory Macro for Microscaled Block-FP in 12-nm CMOS
MANTIS combines mixed-signal computation and microscaled numerical formats, with end-to-end LLM evaluation accounting for hardware non-idealities.
GRU-RVCIM: Efficient Implementation of Gated Recurrent Unit in RISC-V - Compute In Memory co-designed for Edge Applications
Recurrent neural-network inference designed jointly with RISC-V and compute-in-memory hardware for edge applications.
Efficient Closed-Loop Neural Decoding for Motor Control of non-Human Primates with Spiking Neural Networks
Spiking neural networks for efficient closed-loop decoding of motor-control signals.
Beyond papers
Recognition
Our team won both tracks of the IEEE BioCAS 2025 Grand Challenge on closed-loop neural decoding.
Patents
Co-inventor on 5 granted U.S. patents from my work at TCS Research, spanning spike encoding and event-based sensing. View patents
Community
I have reviewed for Nature Communications, Neurocomputing, Neural Networks, CDC, and ICASSP, and co-presented a tutorial on bio-inspired neuromorphic computational paradigms at ICASSP 2023.
Get in touch
I welcome conversations about continual learning, efficient inference, and hardware-aware AI. ageorg23@nd.edu