Rishit Toteja

Microsoft Research

Bangalore, India

I am a Research Fellow at Microsoft Research in the PROSE group, where I work on Copilot for Excel and Formula Intelligence in Excel Spreadsheets. I am advised by Dr. Ashish Tiwari, Dr. Arjun Radhakrishnan, and Dr. Sumit Gulwani.

Previously, I interned twice at Amazon as an Applied Scientist, where I worked on sell rate forecasting for various products across different marketplaces and on developing efficient Text-to-SQL models. I have also interned at Bosch Global Software Technologies as a Software Developer, where I deployed metaverse platforms on immersive devices like Oculus Quest and HoloLens. In addition, I have gained valuable insights from my experiences at the Defence Research and Development Organisation (DRDO) and the Samsung R&D Institute India, where I worked on software development, data science, and AI/ML research.

I completed my B.Tech in Electrical Engineering with a minor in Computer Engineering from Delhi Technological University, graduating with an aggregate GPA of 9.53, ranking 1st in my department.

Outside of work, I am an avid chess and table tennis player.

News (older)

Jul 01, 2024	Joined Microsoft As a Research Fellow with the PROSE Team (led by Sumit Gulwani) working on CoPilot and Formula Intelligence for Excel Spreadsheets.
Jun 28, 2024	Internship ended with Amazon. Got an inclined return FTE Applied Scientist offer.
Mar 23, 2024	Final Project Thesis work (Multivariate Time Series Classification of EEG Signals using Temporal Transformers) accepted in 3rd IEEE International Conference on Power Electronics, Intelligent Control and Energy Systems (ICPEICES).
Mar 07, 2024	Got offer from Microsoft as a research Fellow with the PROSE team.
Jan 11, 2024	Selected as one of the two undergrads from DTU for attending the “Research Week with Google 2024”, a graduate level symposium organised by Google Research India.

Selected Publications (all)

Multivariate Time Series Classification of EEG Signals using Temporal Transformers

Rishit Toteja, Keshav Chauhan, Prince Shishodia, and 3 more authors

In 2024 3rd IEEE International Conference on Power Electronics, Intelligent Control and Energy Systems (ICPEICES), 2024

Abs Bib

In this study, we investigate a novel approach for classifying multivariate time series EEG signals to identify Parkinson’s disease (PD), comparing Transformer models to traditional CNNs. Parkinson’s disease (PD) is a neurodegenerative disorder characterized by progressive loss of dopamine-producing neurons in the brain. Currently, a deep convolutional-recurrent neural network (CRNN) achieves high accuracy in PD detection. We introduce PaCT: a Time Series Transformer with Positional Embeddings (TST-PE), an architecture tailored for EEG signal classification. TST-PE leverages positional embeddings to capture temporal dependencies effectively which embeddings further enhance classification results. Experimental results demonstrate TST-PE’s superior performance over CNN-based approaches, achieving a classification accuracy of 96%. This work suggests Transformer models could significantly improve PD diagnosis, and extends to broader applications in preventive healthcare and brain-computer interface development.
@inproceedings{8897445, author = {Toteja, Rishit and Chauhan, Keshav and Shishodia, Prince and Pillai, Rohan and M. Kumar, Rupesh and Chaudhary, Abhishek}, title = {Multivariate Time Series Classification of EEG Signals using Temporal Transformers}, year = {2024}, volume = {}, number = {}, keywords = {Multivariate Time Series Analysis, Classification, Machine Learning, Transformer, Electroencephalogram (EEG), Parkinson’s Disease}, booktitle = {2024 3rd IEEE International Conference on Power Electronics, Intelligent Control and Energy Systems (ICPEICES)} }
Skin Disease Detection Using Saliency Maps and \nSegmentation Techniques

Rishit Toteja, Dhruv Gupta, Vibhor Gautam, and 1 more author

In International Conference on Computer Vision and Image Processing, 2022

Abs Bib

Skin diseases are among the most prevalent medical conditions globally. Detecting these diseases requires a high degree of expertise and precision from dermatologists, making computer-aided diagnosis highly beneficial. In this paper, we present an approach for detecting skin diseases that provides enhanced diagnostic insights. Several studies have been conducted to aid in the diagnosis of skin conditions, including skin cancer and tumors. However, accurate identification remains a challenge due to factors such as the visual similarity between different conditions, low contrast between lesions and surrounding skin, and the subtle differences between affected and unaffected regions. Our approach employs an image dataset that is pre-processed through deblurring, noise reduction, and enhancement techniques. Following pre-processing, the images are segmented using the U-Net architecture, and MobileNetV2 is employed for classification. The disease is detected based on distinct patterns observed in the processed images, corresponding to the dispersant characteristics of each condition. Additionally, we compute a class saliency map for each input image and its associated class. This work demonstrates how Convolutional Neural Networks (ConvNets) can be used effectively for both the classification and segmentation of skin diseases, utilizing saliency maps to improve interpretability.
@inproceedings{toteja2022skin, title = {Skin Disease Detection Using Saliency Maps and \n Segmentation Techniques}, author = {Toteja, Rishit and Gupta, Dhruv and Gautam, Vibhor and Vishwakarma, Dinesh Kumar}, pages = {328--340}, year = {2022}, organization = {Springer}, booktitle = {International Conference on Computer Vision and Image Processing} }

News (older)

Selected Publications (all)

Current and Past Affiliations