Nichita Uțiu

A paper exploring a novel architecture for extracting text corpus from HTML data.

I developed this as part of my bachelor's thesis, and it is my first ever publication. The architecture achieved state-of-the-art performance at the time, on a known dataset and the performance was thoroughly evaluated. As part of the project, I also deployed the model as a RESTful API and implemented a front end for it using React.

I published a workshop paper at ICML on the viability of using different architectures of graph convolutional networks (GCNs) for few-shot classification. As part of my work as a research assistant in applied machine learning, I explored the possibility of using GCNs to do few-shot classification of nodes on graph-like data such as HTML trees. I also proposed several architectures and empirically tested their performance on existing benchmarks.

A Python library that can be used with PyTorch and Ignite to fine-tune the learning rate of deep learning models using the algorithm proposed in the paper "Cyclical Learning Rates for Training Neural Networks." The library is open-source and well documented.

A detailed summary of Sutton and Barto's Book, "Reinforcement Learning: A second edition." During my work as a research assistant, I investigated RL methods for the project I was working on. In the process, I created a thorough learning aid for the second edition of the seminal RL book that other people reading the book could use.