• What makes Imagenet good for transfer learning?, Minyoung Huh, Pulkit Agrawal, Alexei A. Efros, arXiv 2016 [paper] [website]

  • Pixels to Voxels: Modeling Visual Representation in the Human Brain, Pulkit Agrawal, Dustin Stansbury, Jitendra Malik, Jack Gallant, arXiv 2014 [paper]


  • Curiosity Driven Exploration by Self-Supervised Prediction, Deepak Pathak, Pulkit Agrawal, Alexei A. Efros, Trevor Darrell, International Conference on Machine Learning (ICML) 2017 [paper] [webpage] [video]

  • Learning to Perform Physics Experiments via Deep Reinforcement Learning, Misha Denil, Pulkit Agrawal, Tejas D Kulkarni, Tom Erez, Peter Battalgia, Nando de Freitas, International Conference on Learned Representations (ICLR) 2017 [paper]

  • Combining Self-Supervised Learning and Imitation for Vision based Rope Manipulation, Ashvin Nair*, Dian Chen*, Pulkit Agrawal*, Philip Isola, Pieter Abbeel, Jitendra Malik, Sergey Levine, International Conference on Robotics and Automation (ICRA) 2017 (* equal contribution) [paper] [webpage] [video]

  • Learning to Poke by Poking: Experiential Learning of Intuitive Physics, Pulkit Agrawal*, Ashvin Nair*, Pieter Abbeel, Jitendra Malik, Sergey Levine, Neural Information and Processing Systems (NIPS) 2016 [paper][webpage] (* equal contribution)

  • Generic 3d representation via pose estimation and matching, Amir R Zamir, Tilman Wekel, Pulkit AgrawalColin Wei, Jitendra Malik, Silvio Savarese, European Conference on Computer Vision (ECCV) 2016 [paper]

  • Learning Visual Predictive Models of Physics for Playing Billiards, Katerina Fragkiadaki*, Pulkit Agrawal*, Sergey Levine, Jitendra Malik, arXiv 2015, International Conference on Learned Representations (ICLR) 2016 [paper][preliminary code] (* equal contribution)

  • Human Pose Estimation using Iterative Error Feedback , Joao Carreira, Pulkit Agrawal, Katerina Fragkiadaki, Jitendra Malik, arXiv 2015, Computer Vision and Pattern Recognition (CVPR), 2016 [paper] [code]

  • Learning to See by Moving, Pulkit Agrawal, Joao Carreira, Jitendra Malik, International Conference on Computer Vision (ICCV) 2015 [project]

  • Analyzing the Performance of Multilayer Neural Networks for Object Recognition, Pulkit Agrawal, Ross Girshick, Jitendra Malik, European Conference on Computer Vision (ECCV), 2014 [paper] [sup]

  • The automatic assessment of knowledge interaction processes in project teams, Gahgene Gweon, Pulkit Agrawal, Mikesh Udani, Bhiksha Raj, Carolyn Rose, International Conference of Computer Supported Collaborative Learning (CSCL) 2011 [paper] (Best Student Paper Award)


  • Invariant Object Representation of Images Using Spiking Neural Networks , Pulkit Agrawal, Somdeb Majumdar, US-Patent: 14/228065 [Link]

  • Invariant Object Representation of Images Using Spiking Neural Networks , Pulkit Agrawal, Somdeb Majumdar, Vikram Gupta, US-Patent: 14/228071 [Link]


Computer Vision

Estimation of Improvement in Rosacea using Image Processing [Project Statement] [Project Paper]1 [Code] with Dr. Jonathan Manton (May-July 2009)
This work addresses the need by dermatologists of being able to assess quantitatively the change in severity of the skin disease Rosacea.

Towards Affordance Based Scene Classification [Presentation] [Report] with Dr. Jitendra Malik (Jan-Apr 2012)
This course project reviews some of the classical work in using affordances for object recognition. We propose a framework for scene understanding based on affordances using kinect data.

Scene Analysis and Trajectory Clustering in Surveillance Videos [Report] [Presentation] with Prince Arora, Dr. A. Mukerjee and Dr. K.S. Venkatesh (Jan-Apr 2011)
This work proposes a cognitive setup to measure the similarity of trajectories and presents a computational implementation of the same.

Scene Classification [Mid Term Presentation] [Final Report] [Code] with BVV Srirajdutt, S Nayak and Dr. S Dube (Aug-Nov 2009)
This class project explored identifying scene categories using gabor filters and a tree-based hierarchical classification scheme.

Brain Reading and Computational Neuroscience

A Probe into decoding brain activity using fMRI and MEG [Report] with Dr. Jack Gallant
This work explored the use of MEG brain recordings for the task of predicting visual stimuli presented to subjects. Attempts were also made to jointly model MEG and fMRI data.

Modeling Responses of V1 Neurons to Natural Vision Movies [Report] [Presentation] [Codes]with G Mishra and Dr. A Mukerjee (Aug-Nov 2010)
This class project used method of reverse correlation for estimating spatio temporal receptive fields (STRF) of individual neurons in response to natural vision movies.

Gaver's Hypothesis testing [Report] with Dr. Bhiksha Raj (May-July 2010)
This work looked at developing computational models for validating Gaver's Hypothesis for environmental sound classification.

EEG Data Analysis with Dr. Bhiksha Raj (May-July 2010)
Independent Component Analysis was used to analyze 129 channel EEG data for presence of characteristic activations capable of predicting semantic and temporal coherence in audio-visual stimuli. We were unsuccessful in finding such such cues.

Odd Ones Out!

Automatic assessment of student `reasoning' processes in face-to-face interactions using speech data [Presentation] [Paper] [Code] with G Gweon, M Udani, Dr. B Raj and Dr. C Rose (May-Jul 2010)
We developed computational models capable of identifying time segments when students spoke statements which involved reasoning. This work relies on speech data alone and doesnot employ speech to language conversion.

A Probe Into Dynamic Safe Feature Elimination for LASSO [Report] with Dr. L. Ghaoui (Jan-Apr 2012)
This class project proposes algorithms for incorporating dynamic safe feature elimination within the frameowrk of coordinate descent algorithms used to solve the LASSO problem.

Collocation based approach for training Recurrent Neural Networks [Report] with Dr. P. Abbeel (Aug-Nov 2011)
Training of Recurrent Neural Networks is formulated as a trajectory optimization problem and was implemented using cvx and mosek. However, we were unable to outperform state of art training methods.

System Identification and Filtering with Artificial Neural Networks with Dr. L. Behera (Aug-Nov 2009)
Multilayered networks using back propagation algorithm, dynamic models using Back propogation through time (BPTT) and real time recurrent learning (RTRL) were implemented for system identification. We also explored quantum activation function using Schrodinger wave equation for stochastic filtering of signals.

Robotics, Embedded and Control

Robust Two Degree of Freedom Vehicle Steering Controller Design [Abstract][Results and Code]with Dr. R. Potluri (Jan-Apr 2009)
This class project involved the design of an active steering controller for automobiles in order to stabilize perturbations in dynamics resulting from unexpected disturbances.

Chat Client with P Pandey
Atmega-16 Microcontroller was interfaced with a computer emulating a chat environment involving multiple clients. A PS/2 Keyboard and an LCD screen were interfaced with the MCU to act as input and output units respectively at the client side.

Autonomous Line Following Robot:
Designed an autonomous robot using Atmega-16 MCU which could follow a white curve. The robot employed PD control algorithms, was capable of detection of crossings in a grid of white lines & efficiently took 90 & 180 degree turns. TSOP sensors and 200 rpm dc motors powered by a 12V battery were used.