A Visual Research Journal
A visual record of papers leaving the page through talks, poster sessions, collaborations, and the cities that held them.
From Yokohama and Seattle in 2024 to Barcelona and Seoul in 2026, each stop became part of the same pursuit: making AI lighter, faster, and more reliable.
Contents
The newest field notes appear first.
A year of systems: efficient inference, robust agents, and a summer of research across Los Angeles, Barcelona, and Seoul.
ICML 2026 · COEX
A summer based in Seoul included a visiting scholarship at SNU’s Computing & Memory Architecture Lab and lectures at Samsung Electronics, then converged at ICML week. Posters on pruning, optimization, and robust multi-agent systems placed several branches of the E-AI project in one shared room.
ICASSP 2026 · Main Conference
At ICASSP 2026, “Sharpness-Aware Minimization with Z-Score Gradient Filtering” moved from manuscript to conversation. Supported by a USC GSG Travel Grant, the presentation connected optimization research with the broader signal-processing community, making Barcelona part of the work itself.
USC FoRT-ML
Research is also the community built between deadlines. A spring picnic and a bowling night with USC’s FoRT-ML group capture the quieter infrastructure behind the papers: trust, critique, and time spent together.
Optimization moved to the foreground, from workshop posters in San Diego to a first AAAI gathering in Philadelphia.
NeurIPS 2025 · OPT Workshop
At the NeurIPS OPT workshop, work on low-precision training, Z-score gradient filtering, and auto-scaled normalization met an audience focused on the mechanics of learning itself. The most useful moments happened beside the posters, where assumptions were challenged in real time.
AAAI 2025 · Workshop
The AAAI workshop brought ZNorm, a gradient normalization method for faster skip-connected network training, into a broader AI conversation. Philadelphia supplied the setting: a compact trip between a new research community and the city’s older institutions.
Four stages, four research worlds: optimization, space AI, neural networks, and medical imaging.
IEEE BigData 2024 · BPOD
At the BPOD optimization workshop, the focus was gradient overlap in deep residual networks and how normalization can reshape difficult non-convex training. The final conference trip of 2024 paired a focused technical talk with a winter walk through Washington.
ESA SPAICE 2024
At ESA’s SPAICE conference, AI left the usual benchmark landscape for solar science. The work combined computer vision and ARIMA-LSTM modeling to analyze and predict solar coronal holes. It was a reminder that efficient, reliable learning matters far beyond conventional computing.
IJCNN 2024 · Oral
“Robust Neural Pruning with Gradient Sampling Optimization for Residual Neural Networks” was presented as an oral paper at IJCNN 2024. Yokohama marked an early moment when model-compression work stepped onto an international stage. It was also when a research direction began to look like a sustained program.
CVPR 2024 · CVMI Workshop
At the CVPR CVMI workshop, uncertainty estimation met medical imaging through tumor prediction with unlabeled data. The Seattle trip widened the research lens: reliable AI is not only about efficiency, but also about knowing when a model should hesitate.