Hariharan Manikandan

I am a Masters student in the Machine learning Department at CMU , where I am working with Professor Zico Kolter and Yiding Jiang .

Prior to joining CMU, I worked at Cisco for 2 years as a AI Software Engineer, where my work focused on NLP. I designed and shipped code intelligence products using GPT and encoder-decoder language models. I have also interned at Adobe, San Jose over the summer of 2023. I graduated from VIT Chennai in 2020 with a Bachelors in Computer Science.

Linkedin  /  CV  /  Google Scholar  /  Github

Interests

My interests are spread over NLP (LLMs), multimodal learning, foundation models (CLIP, SAM), and generative modeling (diffusion). I am interested in studying the core ideas behind these models and their applications to real world problems. At a research level, I focus on understanding the theoretical formulation of these models and ways new algorithms can be created to explore their interesting properties.

Only by interacting with a LLM through prompts, one can construct a tabular classifier that can function as weak learner inside boosting.

Multiple-instance learning modeled as an attention-based neural network is able to learn meaningful concepts that can represent various vulnerablities in C code.

A word2vec algorithm for producing semantic-aware embeddings for logs, and feature selection process that identifies the most salient features for detecting anomaly in logs.

Contour features as a weak signal can be useful in supervising segmentation of small-sized lesions in CT scans of COVID-19 patients.

Inducing an optimization scheme that restricts the maximal activation of convolutional filters (in a specific layer) towards a certain target class.

Using spatial attention over tomato leaf images detects the presence of common diseases with high accuracy.

This page was created from Jon barron's template.