Ph.D. Candidate · Since 2022
Deep Learning & Continual Learning Researcher
Department of Data Science and Engineering,
IISER Bhopal · Supervisor: Prof. Vinod K Kurmi
I build vision models that keep learning from a live data stream — absorbing new classes as they arrive without forgetting the ones they already know, and without storing old data or being told where one task ends and the next begins.
IISER Bhopal · Madhya Pradesh, India
Research Interests
Every thread below returns to one question: what must stay fixed, and what is allowed to move, when a model meets a class it has never seen before?
Each new class arrives with only a handful of samples, so almost everything rides on how good the base representation was beforehand. I shape that base feature space through feature augmentation, leaving room for classes it was never trained on.
Real data does not arrive in tidy labelled tasks. It arrives once, in a single pass, with classes quietly mixing and shifting. GOFOR handles this with no gradient updates at all — an orthogonal random projection, prototypes separated using the running covariance, and a drift detector that decides when the model may change.
Vision is not the only stream that keeps growing. I work on class-incremental learning where audio and video must be reconciled, and on speech recognition that adapts across Indian languages instead of being retrained from scratch for each one.
Each cluster is one of my papers, arriving in publication order. Nothing that arrived earlier is pushed aside — that is exactly the problem continual learning has to solve.
Publications
Four papers across computer vision, speech and applied AI venues — two of them accepted in 2026.
Parinita Nema, Vinod K Kurmi
Organises the feature space through closed-form updates instead of backpropagation, stores no exemplars, and refreshes only when a Sliced-Wasserstein drift detector confirms the stream has actually moved. Improves on prior work by up to 14.5% across CIFAR-10, CIFAR-100, CORe50 and CUB-200.
Rini Smita Thakur, Aditya Kumar Singh, Parinita Nema, Vinod Kumar Kurmi
Feature-guided prompting for class-incremental learning when the arriving classes are audio-visual rather than image-only.
Parinita Nema, Vinod K Kurmi
Builds a base representation that leaves usable room for classes it has not met yet, so the few-shot sessions that follow are not fighting the backbone.
Riyansha Singh, Parinita Nema, Vinod K Kurmi
Carries the few-shot incremental setting over to audio, using contrastive representations to keep earlier sound classes separable.
Achievements
August 2026 · Research Grant
Awarded by the Innovation & Incubation Centre for Entrepreneurship (IICE), IISER Bhopal, to develop VaaniAdapt — a continual learning platform for multilingual Indian-language speech recognition.
July 2026
Symposium on Energy and AI, IISER Bhopal — for Continual Learning for Smart Building Energy Forecasting.
2025
Awarded to support attendance and presentation at WACV 2025.
2021
UGC National Eligibility Test — Junior Research Fellowship, Computer Science and Applications.
2020
Outstanding Student Performance, Vikram University, Ujjain.
2023
Selected participant, Computer Vision summer school at IIIT Hyderabad.
2023
Healthcare services and affective computing using deep learning, IIT Indore.
Projects
Taking continual learning out of the benchmark and pointing it at problems with a user on the other end.
Funded · SIG 3.0 · 2026–present
Continual learning for multilingual speech recognition
A speech recognition platform that adapts across Indian languages as it goes, rather than being retrained from scratch for every new one. Supported by a Student Innovation Grant from IICE, IISER Bhopal.
Award-winning · 2026
Continual learning on non-stationary energy data
Building energy demand shifts with seasons, occupancy and equipment — the same non-stationarity problem, in a setting where being wrong wastes power. Presented at the Symposium on Energy and AI, IISER Bhopal.
Teaching
Six semesters at IISER Bhopal, across the introductory programming course and both deep learning courses.
Teaching Assistant · IISER Bhopal
Teaching Assistant · IISER Bhopal
Teaching Assistant · IISER Bhopal
Teaching Assistant · IISER Bhopal
Teaching Assistant · IISER Bhopal
Teaching Assistant · IISER Bhopal
Education
IISER Bhopal · Thesis: Incremental Learning in Computer Vision · Supervisor: Prof. Vinod K Kurmi · CPI 8.5/10
Vikram University, Ujjain · 82.58%
Vikram University, Ujjain · 72.30%
Post Graduate Diploma in Computer Applications
Where I work
Indian Institute of Science Education and Research Bhopal, Bhauri, Madhya Pradesh.
Contact
I am always glad to discuss continual learning, and open to collaborations, internships and reviewing. Department of Data Science and Engineering, IISER Bhopal, Bhauri, Madhya Pradesh 462066, India.