Akshita Ramya Kamsali

185, BHEE

465 Northwestern Ave

West Lafayette, IN

I am a PhD candidate in Electrical and Computer Engineering at Purdue University and a member of the Robot Vision Lab, under the guidance of Dr. Avi Kak. My research lies at the intersection of Natural Language Processing (NLP) and Computer Vision (CV), focusing on machine learning solutions for multimodal data analysis, with expertise in LLM evaluation, knowledge graph generation, inference, and anomaly detection [NeurIPS’24]. Proficient in Python, PyTorch, and Deep Learning, I also bring strong skills in scientific communication, public speaking, and presentations, enabling me to effectively bridge research and impactful real-world applications.

For a detailed overview of my academic and professional journey, you can view my full CV or my condensed resume.

Let’s connect and explore how technology can transform the way we perceive and solve problems!

News

Dec 10, 2024	NeurIPS 2024 - Received NSF travel grant to present poster on my change detection work.
Oct 10, 2024	Chair for ECE Graduate Symposium 2024. Moderated panel discussion on “ECE in the Age of AI”
Nov 30, 2023	Appointed GTA for ECE60146: Deep Learning
Apr 12, 2023	Elected President for ECEGSA and Career Chair, PGSG
Jul 22, 2022	Starting as member of RVL and instructor for ECE20007

Selected Publications

IEEE-WRAP

Experimental Verification of Enhanced Photoluminescence in p-doped GaAs using Fluorescence Lifetime Measurements

Ramya, Kamsali Akshita, Jinal, Tapar, Saurabh, Kishen, and Emani, Naresh Kumar

In 2019 Workshop on Recent Advances in Photonics (WRAP) Dec 2019

Abs HTML

The realization of low threshold, room temperature operation of semiconductor NIR nanolasers is still a challenge. Large surface-to-volume ratio of the nanostructures aggravates surface recombination leading to non-radiative decay of excited carriers. P-doping of active material is a viable solution to bring down the lasing threshold without surface passivation. Here, we report an improved quantum yield of p-doped GaAs over intrinsic, with the help of carrier lifetime measurement using FLIM system. We aim to study how the onset of the Auger process affects the PL and carrier lifetime at higher pump fluence.
NeurIPS-WiML

Graph-based Change Detection for Natural Language

Kamsali, Akshita Ramya

In Dec 2024

Abs HTML

Computational models such as machine learning (ML) and deep learning (DL) excel at processing structured data, including numbers, images, audio signals, and spectral data. These types of data can be easily converted into numerical patterns, which models can analyze to extract meaningful insights. In contrast, processing text is significantly more complex. Language is inherently rich, intricate, and nuanced. Techniques like tokenization and word embeddings, which transform text into numerical representations, often struggle to fully capture the subtle context and deeper meaning within the text. Challenges such as context dependency, disambiguation, and diverse writing styles can lead to multiple interpretations of the same text, even for humans. This complexity is further magnified in specialized domains, such as legal or medical documents. In this work we present LLM-based solution to build consistent knowledge graphs for change detection.