Kshitij Dwivedi

I am a PhD student at CVAI lab in Goethe University Frankfurt, where I am advised by Prof. Gemma Roig.

Before starting my PhD I've worked as an Engineer at Kamitani lab in ATR, Japan and Samsung R&D India. I did my Bachelors and Masters in Electrical Engineering at IIT Kanpur.

Email  /  CV  /  Publications  /  Google Scholar  /  LinkedIn


Research Interests

I am interested in understanding how the human visual cortex works and then apply the gained insights towards creating agents with lifelong learning capabilities. I collaborate with Radek Cichy's lab towards achieving my research goal. Below you can find my publications related to Computer Vision and Human Vision highlighted in respective colors.


Publications
Unveiling functions of the visual cortex using task-specific deep neural networks
Kshitij Dwivedi, Michael F. Bonner, Radoslaw Martin Cichy*, Gemma Roig*
PLOS Computational Biology, 2021 (in press) 
* jointly directed work
Project Page / preprint / Code / CCN 2019

We investigated the potential of deep neural networks trained on a diverse set of tasks in finding functionals roles of different regions of the human visual cortex

The Algonauts Project 2021 Challenge: How the Human Brain Makes Sense of a World in Motion
Radoslaw Martin Cichy, Kshitij Dwivedi, Benjamin Lahner, Alex Lascelles, Polina Iamshchinina, Monika Graumann, Alex Andonian, N Apurva Ratan Murty, Kendrick Kay, Gemma Roig, Aude Oliva
arxiv, 2021  

Project Page / arxiv / Code

We present Algonauts 2021 challenge where the goal is to determine which computational model best explains human brain responses while humans view everyday events

Duality Diagram Similarity: a generic framework for initialization selection in task transfer learning
Kshitij Dwivedi, Jiahui Huang, Radoslaw Martin Cichy, Gemma Roig
ECCV, 2020
pdf / code

Highly efficient and accurate approach for initialization selection in task transfer learning.

Unravelling Representations in Scene-selective Brain Regions Using Scene Parsing Deep Neural Networks
Kshitij Dwivedi, Radoslaw Martin Cichy*, Gemma Roig*
Journal of Cognitive Neuroscience, 2020  
*jointly directed work
Early Access / bioRxiv / ECCV workshop 2018

We show that a scene parsing model explains human scene-selective responses better than a scene classification model. Further, components predicted by scene parsing models can be used to distinguish functional roles of human scene-selective areas PPA and OPA.

Representation Similarity Analysis for Efficient Task taxonomy & Transfer Learning
Kshitij Dwivedi, Gemma Roig
CVPR, 2019
pdf / code

Efficient method to estimate task similarities and its application to transfer learning

End-to-End Deep Image Reconstruction From Human Brain Activity
Guohua Shen*, Kshitij Dwivedi*, Kei Majima , Tomoyasu Horikawa, Yukiyasu Kamitani
Frontiers in Computational Neuroscience, 2019
*equal contribution
code

Reconstructing perceived images from human fMRI responses.

The Algonauts Project: A Platform for Communication between the Sciences of Biological and Artificial Intelligence
Radoslaw Martin Cichy, Gemma Roig, Alex Andonian, Kshitij Dwivedi, Benjamin Lahner, Alex Lascelles, Yalda Mohsenzadeh, Kandan Ramakrishnan, Aude Oliva
Conference on Cognitive Computational Neuroscience (CCN) , 2019
Workshop

We organized a challenge and workshop to predict human fMRI and MEG responses using computational models.


Conference abstracts and Workshop Papers
Unveiling functions of visual cortex using task-specific deep neural networks
Kshitij Dwivedi, Michael F. Bonner, Radoslaw Martin Cichy, Gemma Roig
Neuromatch 2.0 , 2020   (Short talk)
slides

Deep Anchored Convolutional Neural Networks
Jiahui Huang, Kshitij Dwivedi, Gemma Roig
Computer Vision and Pattern Recognition Workshop (CVPRW) on Compact and Efficient Feature Representation and Learning (CEFRL), 2019   (Oral Presentation)
pdf

Neural network compression using convolutional parameters sharing.

Explaining Scene-selective Visual Area Using Task-specific and Category Specific DNN Units
Kshitij Dwivedi, Michael F. Bonner, Gemma Roig
Vision Sciences Society Conference, 2019
Abstract

Navigational Affordance Cortical Responses Explained by Semantic Segmentation model
Kshitij Dwivedi, Gemma Roig
European Conference on Computer Vision Workshops (ECCVW) on Brain-Driven Computer Vision (BDCV), 2018
pdf

Relating functions of visual cortex to deep neural network functions.

Plug and Play DNN Modules for Multi-domain Learning
Kshitij Dwivedi, Gemma Roig
European Conference on Computer Vision Workshops (ECCVW) on Interactive and Adaptive Learning in an Open World (IAL), 2018
pdf

Efficient multi-domain learning using parameter sharing.

Importance of object selection in Relational Reasoning tasks
Kshitij Dwivedi, Gemma Roig
Neural Information Processing Systems Workshops on Relation Representation Learning (R2L), 2018
pdf


Teaching
TA, Computer Vision 2020-21, Goethe University Frankfurt am Main
TA, Multivariate EEG online school 2020
Lead TA, Neuromatch Academy 2020
TA, Data Structure and Algorithms 2014, IIT Kanpur

I borrowed the template from this great homepage