A Look Into My Research Journey

Quick update: I now have two full papers accepted at IEEE EMBC 2026. I wanted this post to be a short, human summary of what they are about while I finish the full publication pages.

Paper 1

Kolmogorov-Arnold Networks for Alzheimer's Disease Detection using Multimodal Speech Analysis

This paper focuses on early Alzheimer's screening from speech and transcripts. The core idea was practical: instead of fully fine-tuning huge models, we kept large pre-trained audio/text encoders frozen and trained lightweight classifier heads on top. That makes the approach more data-efficient and compute-efficient for low-data settings.

We compared three head designs (MLP, pure KAN, and a Feedforward-KAN-Feedforward setup). The FKF head gave the most stable results, and the best multimodal late-fusion setup reached 0.8556 ± 0.0382 macro-F1 and 0.9220 ± 0.0295 AUC. We also included mechanistic analyses to understand why the architecture works, not just whether it works.

Paper 2

Analyzing Coronavirus Cough Sounds and Alzheimer's Speech with Deep Learning Models to Discover Common Patterns

This paper explores whether COVID-19 and Alzheimer's data can show overlapping acoustic/linguistic patterns. We combined ADReSSo speech data with crowdsourced COVID-19 cough recordings, used deep-learning embeddings, and examined clustering behavior across cohorts.

One interesting finding was pattern-level proximity between embeddings from younger COVID-19 participants and older Alzheimer's participants across multiple encodings. We also benchmarked Alzheimer's classification from 2D speech representations: Inception-v3 + MFCC performed best at around 70% accuracy/macro-F1, and ensembles improved this to roughly 73% accuracy and 72% macro-F1.

What is coming next

The publication entries are already on my site, but the detailed publication pages and official proceedings/PDF links are still being finalized. I will publish those as soon as they are available.