Dipan Pal


I am third year ECE PhD student at Carnegie Mellon University. I work with Prof. Marios Savvides, at the CyLab Biometrics Center. I also occasionally collaborate with Prof. Ole Menshoel. My current interests lie in the intersection of computer vision and machine learning.

Most of my work is centered around the problem of perception and representation learning. Building useful representations of data is of paramount importance. Ideally, a single good representation would be useful for any task one is interested in. Towards this goal, representations that are invariant to nuisance within-class transformations and that are discriminative towards between class transformations are very useful. Currently I am modelling nuisance transformations as unitary, as inspired by the work of Prof. Tomaso Poggio. Moving forward, for my theoretical thrust, I am starting to investigate models which utilize hierarchical non-linear compositions of unitary transformations and whether they can approximate the complicated non-linear transformations that appear in real-world data. Along similar lines as a practical thrust, I am also working to introduce a new class of feed forward networks called Transformation Networks. Transformation Networks can be of Parametric and Non-parametric type. Convolutional Neural Networks are a type of Parametric Transformation Networks. However, a more general and (potentially) more powerful network would be of the non-parametric type, leading to Non-parametric Transformation Networks (NPTNs).

Moving forward, I am broadly interested in understanding how perception can help reasoning and vice versa? what role does feedback play in these systems? How do you have an agent seamlessly integrate their vision systems with higher cognition? Further, the paradigm ofdeep reinforcement learning seems intriguing while I was sitting in on Deep Reinforcement Learning at CMU by Ruslan and Katerina. A lot of open problems there.

In the works

Dipan K. Pal and Marios Savvides, Non-parametric Tranformation Networks
Dipan K. Pal and Marios Savvides, General Perception Models
Dipan K. Pal and Marios Savvides, Planet Loss for Face Recognition

In the pipeline

Dipan K. Pal and Marios Savvides, How Transformation Networks model Non-linear Invariances (under review at ICML 2017) preprint
Ngan Le, Chi Nhan Duong, Ligong Han, Khoa Luu, Marios Savvides, Dipan K. Pal, Deep Contextual Recurrent Residual Networks for Scene Labeling (under review at ICML 2017)
Dipan K. Pal, Ashwin A. Kannan*, Gautam Arakalgud* and Marios Savvides, Max-margin Invariant Features: Unitary-Group Invariant Kernels from Transformed Unlabeled Data (under review at CVPR 2017, * indicates equal contribution)
Pokkalla Harsha Vardhan, Kunal Sekhri, Dipan K. Pal and Marios Savvides, Class Correlation affects Single Object Localization using Pre-trained ConvNets (under review at WACV 2017)
Shreyas Venugopalan, Dipan K. Pal, Marios Savvides, Exploiting Sparsity in Local Iris Texture Representation: An Iris Spoofing Paradigm (under review)

Publications (major contribution)

Dipan K. Pal, Vishnu N. Boddeti and Marios Savvides, Emergence of Selective Invariance for Object Categorization through Adaptive Pooling , arXiv, 2017

Dipan K. Pal and Marios Savvides, Unitary-Group Invariant Kernels and Features from Transformed Unlabeled Data, arXiv, 2016

Dipan K. Pal and Ole Mengshoel, Stochastic CoSaMP: Randomizing Greedy Pursuit for Sparse Signal Recovery,, ECML-PKDD, 2016. Supplementary Material

Dipan K. Pal, Felix Juefei-Xu and Marios Savvides, Discriminative Invariant Kernel Features: A Bells-and-Whistles-Free Approach to Unsupervised Face Recognition and Pose Estimation, CVPR 2016 (spotlight presentation)

Felix Juefei-Xu, Dipan K. Pal and Marios Savvides, Hallucinating the Full Face from the Periocular Region via Dimensionally Weighted K-SVD, CVPRW Biometrics, 2014

Dipan K Pal, S. S. Tripathy, Viresh Ranjan, Avinash Das, Robust Content Based Image Retrieval system using Hierarchical Temporal Memory, ICECT, 2012 (only paper proceedings published)

Dipan K Pal, Evaluation of Gabor Filter enhanced Hierarchical Temporal Memories in image classification applications, IEEE ICCIC, 2011 (only paper proceedings published)

Dipan K Pal, Biometric identification using Hierarchical Temporal Memory with face recognition as a case study, IEEE ICCIC, 2011 (only paper proceedings published)

Other Publications

Felix Juefei-Xu, Dipan K. Pal and Marios Savvides, NIR-VIS Heterogeneous Face Recognition via Cross-Spectral Joint Dictionary Learning and Reconstruction, CVPRW Perception Beyond the Visual Spectrum Workshop, 2015

Keshav Seshadri, Felix Juefei-Xu, Dipan K. Pal, Marios Savvides and C.P. Thor, Driver Cell Phone Usage Detection on Strategic Highway Research Program (SHRP2) Face View Videos, CVPRW Workshop on Computer Vision in Vehicle Technology, 2015

Felix Juefei-Xu, Dipan K. Pal, Karanhaar Singh, and Marios Savvides, A Preliminary Investigation on the Sensitivity of COTS Face Recognition Systems to Forensic Analyst-style Face Processing for Occlusions, CVPRW Biometrics, 2015

Niv Zehngut, Felix Juefei-Xu, Rishabh Bardia, Dipan K. Pal, Chandrasekhar Bhagavatula, and Marios Savvides, Investigating the Feasibility of Image-Based Nose Biometrics, ICIP, 2015

Jun Shi, Ole Mengshoel, Dipan K. Pal, Feedback control for multi-modal optimization using genetic algorithms, GECCO, 2014


Felix Juefei-Xu, Dipan K. Pal, and Marios Savvides, “Methods and Software for Hallucinating Facial Features by Prioritizing Reconstruction Errors”, U.S. Provisional Patent Application Serial No. 61/998,043, June 17, 2014.


Please download a copy of my resume from here.