Deep Learning  & AI 

Overview

Deep Learning(DL) has propelled rapid advances in AI over the past decade.  However, applying DL techniques to real world enterprise problems is far from straightforward. Major challenges include missing or noisy training data, changing privacy regulations, heavy compute and memory requirements,  lack of interpretability and principled techniques for imbibing causal information and explicit domain knowledge into DL models. 

Research Area: Led by Senior Scientist Dr. Lovekesh Vig,

The DLAI research area at TCS Research aims to address these challenges and further the quest to develop robust, readily deployable enterprise AI solutions. To that end we focus on the following research themes:

1. Deep Meta Learning: This group works on developing novel techniques for deep meta-learning, transfer learning and domain adaptation, with a view towards applications in industrial IoT, time series forecasting, and Personalization. 

Head: Pankaj Malhotra, The team: 

2. Deep Visual Intelligence: The team works on combining deep learning and traditional vision techniques to develop real world solutions to Enterprise vision problems. While deep learning models often yield state of the art results for detection and recognition tasks, the information has to be combined with explicit prior knowledge and neuro-symbolic techniques need to be developed on top of the deep learning predictions to solve the enterprise problem.The current application areas  in focus are information extraction from document images and medical image processing. 

Head: Monika Sharma, 

Asset: DeepReader

3. Causal Inference:

For machine learning to be truly effective in a generalized setting, machine learning models need to understand the difference between correlation and causation, be able to discover the underlying causal structure for a problem and utilize causal information while making predictions. This group  works on developing techniques  for estimating treatment effect for high dimensional and temporal treatment scenarios, and causal reinforcement learning with a view towards applications in online advertising and healthcare.

Application Areas: Online Advertising, Healthcare

Head: Garima Gupta, 

4. Next-Generation BI

This team  works on the next generation of BI solutions that will allow for seamless interaction with complex structured data. This includes conversational systems for performing standard BI tasks such as generation of reports and visualizations and performing complex queries in addition to generating textual summaries of tabular data

Application Area: Healthcare, Energy

Team Lead: Dr. Amit Sangroya, The team:

5. Resource Efficient ML

With deep models getting larger and deeper, there is an increasing need to develop techniques to reduce the memory and compute footprint of these models, especially for on device applications for robotics, augmented/virtual reality and others. This team works on techniques for model compression, on-device 3D vision and 

Application Area: Manufacturing, IoT

Team Lead: Ramya Hebbalaguppe, 

6.Learning under Data Sparsity 

While deep learning has performed admirably for pattern recognition tasks like image recognition, principled techniques for incorporating   reasoning and explicit domain knowledge remain elusive. This team works on combining the pattern recognition capabilities of deep learning with the reasoning capabilities of traditional symbolic techniques via neuro-symbolic integration, high level reasoning and human in the loop AI. The primary application areas are towards understanding scientific documents and online advertising.

Head: Dr. Shirish Karande, 

Assets: Samuhaa platform for human in the loop AI

 7. NLP for Conversational Systems

Despite the recent breakthroughs using self supervised learning with Transformer architectures for generating contextual embeddings for words, sentences and paragraphs, there is still a wide usability gap while trying to conduct general conversations with a conversational agent. Most chatbot applications are still quite narrow. The NLP team is building techniques to imbibe conversational agents with the ability to answer out of domain questions, automate business processes, imbibe and enhance knowledge via user/web interactions, reduce dependence on training data for adaptation to novel domains/languages and eventually become an evolving repository of enterprise knowledge. 

Head: Mayur Patidar, The team:

Assets: KNADIA

Academic partners: BITS Goa, India; and IIT Delhi, India

Publications and Patents 

Journals:

Narendhar Gugulothu, Vishnu Tv, Pankaj Malhotra, Lovekesh Vig, Puneet Agarwal, and Gautam Shroff, "Predicting Remaining Useful Life Using Time Series Embeddings based on recurrent neural networks", International Journal of Prognostics and Health Management (IJPHM), 2018. 

Priyanka Gupta, Vishnu TV, Pankaj Malhotra, Lovekesh Vig, and Gautam Shroff, Transfer Learning for Clinical Time Series Analysis using Deep Neural Networks, Springer Journal of Healthcare Informatics , 2019

Ashwin Srinivasan, Lovekesh Vig and Michael Bain, Logical Explanations for Deep Relational Machines Using Relevance Information, Journal of Machine Learning Research (JMLR), 2019

Ashwin Srinivasan, Lovekesh Vig and Gautam Shroff, Constructing generative logical models for optimisation problems using domain knowledge, Machine Learning Journal (MLJ), 2019 

Conferences

1. Pankaj Malhotra, Anusha Ramakrishnan, Gaurangi Anand, Lovekesh Vig, Puneet Agarwal, Gautam Shroff, "LSTM-based Encoder-Decoder for Multi-sensor Anomaly Detection", ICML Workshop on Anomaly Detection, 2016, New York
2. Pankaj Malhotra, Vishnu TV, Anusha Ramakrishnan, Gaurangi Anand, Lovekesh Vig, Puneet Agarwal, Gautam Shroff
3. "Multi-Sensor Prognostics using an Unsupervised Health Index based on LSTM Encoder-Decoder", KDD Workshop on ML for Prognostics and Health Management, 2016, San Francisco
4. Ashwin Srinivasan, Gautam Shroff, Lovekesh Vig, Sarmila Saikia, Puneet Agarwal, "Generation of Near Optimal Solutions using ILP guided sampling" ,(* Best Paper Award) ILP 2016, London 
5. Perla Ramakrishna, Ramya Hebbalaguppe, Gaurav Gupta, Geetika Sharma, Ehtesham Hassan, Monika Sharma, Lovekesh Vig, Gautam Shroff. "An AR Inspection Framework: Feasibility Study with Multiple AR Devices", ISMAR, Merida, 2016 
6. Ankit Verma, Monika Sharma, Ramya Hebbalaguppe, Ehtesham Hassan, Lovekesh Vig, "Automatic Container Code Recognition via Spatial Transformer Networks and Connected Component Region Proposals", ICMLA, Anaheim, 2016
7. Anirban Chakraborti, Kiran Sharma, Aditeya Pandey, Kaushal Paneri, Sidharth Verma, Gunjan Sehgal, Bindu Gupta, Geetika Sharma, Lovekesh Vig, Puneet Agarwal, Gautam Shroff, "Spatio-Temporal Analysis of Ethnic Conflicts and Human Rights Violations in Africa and Middle East", SBP-BRIMS, Washington DC, 2016
8. Sarmimala Saikia, Lovekesh Vig, Ashwin Srinivasan, Gautam Shroff, Puneet Agarwal, Richa Rawat, " Knowlege-Rich Deep Networks for Optimization by Inclusion of ILP-constructed Features ", NIPS Workshop on Cognitive Computing and Neuro Symbolic Integration, Barcelona, 2016
9. Karamjit Singh, Garima Gupta, Lovekesh Vig, Gautam Shroff, and Puneet Agarwal, "Deep Convolutional Neural Networks for Pairwise Causality", What If? NIPS workshop on on hypothetical and counterfactual interventions, Barcelona, 2016
10. Mohit Yadav, Lovekesh Vig, and Gautam Shroff, "Learning and Knowledge Transfer with Memory Networks for Machine Comprehension", EACL, Valencia, 2017
11. Pankaj Malhotra, Lovekesh Vig, Puneet Agarwal, and Gautam Shroff, "TimeNet: Using Pre-trained Embeddings based on LSTM-Autoencoders for Time-Series Classification and Clustering", ESANN, Bruges, 2017 
12. Monika Sharma, Oindrila Saha, Anand Sriram, Shirish Karande,and Lovekesh Vig, "Crowdsourcing for Chromosome Segmentation and Deep Classification", CVMI Workshop, CVPR, Hawaii, 2017
13. Narendhar Gugulothu, Vishnu Tv, Pankaj Malhotra, Puneet Agarwal, Lovekesh Vig, and Gautam Shroff "Predicting Remaining Useful Life using Time Series Embeddings based on Recurrent Neural Networks", 2nd ACM SIGKDD Workshop on Machine Learning for Prognostics and Health Management, Halifax, 2017
14. Vishnu Tv, NarendharGugulothu, Pankaj Malhotra, Puneet Agarwal, Lovekesh Vig and Gautam Shroff, Explainable Deep Learning for Health Monitoring of Complex Systems AI4IOT Workshop, IJCAI, Melbourne, 2017
15. Gaurav Gupta, Swati J, Monika Sharma, and Lovekesh Vig, Data Extraction from Handmarked Inspection Sheets Workshop on Camera Based Document Analysis and Recognition, ICDAR, Kyoto, 2017
16. Lovekesh Vig, Ashwin Srinivasan, Michael Bain, and Ankit Verma, "An Investigation into the Role of Domain-Knowledge on the Use of Embeddings", ILP, Orleans, 2017
17. Ashwin Srinivasan, Lovekesh Vig and Gautam Shroff, "Mode-Directed Neural-Symbolic Modelling" , ILP, Orleans, 2017
18. Swati J, Gaurav Gupta, Mohit Yadav, Monika Sharma, and Lovekesh Vig, Siamese Networks for Chromosome Classification, Bioimage Computing workshop, ICCV, Venice 2017
19. Somdyuti Paul and Lovekesh Vig, "Deterministic Policy Gradient Based Robotic Path Planning with Continuous Action Spaces", Workshop on Vision in Practice on Autonomous Robotics, ICCV , Venice, 2017
20. Mayur Patidar, Puneet Agarwal, Lovekesh Vig, Gautam Shroff, "Correcting Linguistic Training Bias in an FAQ-bot using LSTM-VAE", ECML-PKDD Data Mining and Natural Language Processing (DMNLP) Workshop, Macedonia, 2017
21. Prerna Khurana, Puneet Agarwal, Ashwin Srinivasan, Gautam Shroff, and Lovekesh Vig, "Hybrid BiLSTM-Siamese network for FAQ Assistance", CIKM Case Studies Track,Singapore 2017
22. S Vishal, Mohit Yadav, Lovekesh Vig and Gautam Shroff, "Information Bottleneck inspired method for chat text segmentation", IJCNLP, Taiwan, 2017.
23. D. Vishwanath, Lovekesh Vig, Puneet Agarwal and Gautam Shroff, "MEETING BOT: Reinforcement Learning for Dialogue Based Meeting Scheduling", AAAI 2018, DEEPDIAL Workshop on Reasoning and Learning for Human-Machine Dialogues, New Orleans, 2018.
24. Sakti Saurav, Pankaj Malhotra, Vishnu Tv, Narendhar Gugulothu, Lovekesh Vig, Puneet Agarwal, Gautam Shroff, "Online Anomaly Detection with Concept Drift Adaptation using Recurrent neural Networks", IKDD, Conference on Data Science (CoDS), Goa, 2018.
25. Prerna Khurana, Puneet Agarwal, Gautam Shroff, Lovekesh Vig, "Resolving Abstract Anaphora Implicitly in Conversational Assistants using a Hierarchically stacked RNN", KDD (Applications Track), London, 2018
26. Monika Sharma, Swati J, Lovekesh Vig, "Automatic Chromosome Classification using Deep Attention Based Sequence Learning of Chromosome Bands", IJCNN, Rio, 2018
27. S. Vishal, Arnab Chatterjee, Lovekesh Vig, Gautam Shroff, "Prosocial or Selfish?: Multi-Behaviour Agents for Contract Negotiation using Reinforcement Learning", IJCAI Workshop on Automated Negotiating, Stockholm, 2018
28. Priyanka Gupta, Pankaj Malhotra, Lovekesh Vig, and Gautam Shroff, "Transfer Learning for Clinical Time Series Analysis using Recurrent Neural Networks", IJCAI Workshop in Knowledge Discovery in Healthcare, Stockholm, 2018
29. Priyanka Gupta, Pankaj Malhotra, Lovekesh Vig, and Gautam Shroff, Using Features From Pre-trained TimeNET For Clinical Predictions, IJCAI Workshop in Knowledge Discovery in Healthcare, Stockholm, 2018
30. Narendhar Gugulothu, Pankaj Malhotra, Lovekesh Vig and Gautam Shroff, "Sparse Neural Networks for Anomlay Detection in High Dimensional Time Series", IJCAI Workshop on AI4IOT, Stockholm, 2018 
31. Vishnu TV, Pankaj Malhotra, Lovekesh Vig, Gautam Shroff, "Deep Ordinal Regression for Remaining Useful Estimation from Censored Data", ICML Workshop on Deep Learning for Safety-Critical Applications in Engineering Systems, Stockholm, 2018
32. Arindam Chowdhury and Lovekesh Vig, "An efficient end-to-end neural model for handwritten text recognition ", BMVC, 2018
33. Swati J, Monika Sharma, and Lovekesh Vig, "Automatic Classification of Low Resolution Classification Images", ECCV Workshop on Bio-Image Computing, Munich, 2018
34. Monika Sharma; Abhishek Vermaa; Lovekesh Vig, "Learning to Clean: A GAN Perspective", International Workshop on Robust Reading, ACCV, Perth, 2018 
35. Rohit Rahul; Arindam Chowdhury; Animesh Animesh; Samarth Mittal; Lovekesh Vig, "Reading Industrial Inspection Sheets by Inferring Visual Relations", International Workshop on Robust Reading, ACCV, Perth, 2018 
36. Vishwanath D; Rohit Rahul; Gunjan Sehgal; Swati; Arindam Chowdhury; Monika Sharma; Lovekesh Vig; Gautam Shroff; Ashwin Srinivasan, "Deep-Reader: End-to-end framework for relevant text extraction from real image documents with Natural Language Interface", International Workshop on Robust Reading, ACCV , Perth, 2018
37. Tirthraj Dash, Ashwin Srinivasan, Lovekesh Vig, Oghenejokpeme I. Orhobor, and Ross King Large-Scale Assessment of Deep Relational Machines, 28th International Conference on ILP 2018, Ferrara, Italy, 2018,***Best Student Paper
38. Mayur Patidar, Puneet Agarwal, Lovekesh Vig, and Gautam Shroff, Automatic Conversational Helpdesk Solution using Seq2Seq and Slot-filling Models., CIKM, Turin, 2018
39. Sarmimala Saikia, Richa Verma, Puneet Agarwal, Gautam Shroff, Lovekesh Vig and Ashwin Srinivasan, Evolutionary RL for Container Loading, ESAAN, Bruges, 2018
40. Arijit Ukil, Pankaj Malhotra, Soma Bandyopadhyay, Tulika Bose, Ishan Sahu, Ayan Mukherjee, Lovekesh Vig, Arpan Pal, and Gautam Shroff, Fusing Features based on Signal Properties and TimeNet for Time Series Classification, ESAAN, Bruges, 2019
41. Kathan Kashiparekh, Jyoti Narwariya, Pankaj Malhotra, Lovekesh Vig, Gautam Shroff, ConvTimeNet: A Pre-trained Deep Convolutional Neural Network for Time Series Classification , IJCNN, Budapest, 2019
42. Monika Sharma, Shikha Gupta, Arindam Chowdhury and Lovekesh Vig. ChartNet: Visual Reasoning over Statistical Charts using MAC-Networks, IJCNN, Budapest, 2019
43. Saurabh Srivastava, Puneet Agarwal, Gautam Shroff, Lovekesh Vig, Hierarchical Capsule Based Neural Network Architecture for Sequence Labeling, IJCNN, Budapest, 2019
44. Diksha Garg, Priyanka Gupta, Pankaj Malhotra, Lovekesh Vig and Gautam Shroff, Sequence and Time Aware Neighborhood for Session-based Recommendations, SIGIR, Paris, 2019 
45. Ankit Sharma, Garima Gupta, Ranjitha Prasad, Arnab Chatterjee, Lovekesh Vig, Gautam Shroff , "MetaCI: Meta-Learning for Causal Inference in a Heterogeneous Population", Workshop on Do the right thing”: machine learning and causal inference for improved decision making, Neurips 2019, Montreal
46. Vishnu T. V. , Pankaj Malhotra, Jyoti Narwariya, Lovekesh Vig, Gautam Shroff, Meta-Learning for Black Box Optimization, ECML 2019 (Main Track)
47. Jyoti Narwariya, Pankaj Malhotra, Lovekesh Vig, Gautam Shroff, T. V. Vishnu, Meta-Learning for Few-Shot Time Series Classification, CoDS, 2020 (Main Track)
48. Shubham Palliwal, Vishwanath D, Rohit Rahul, Monika Sharma, Lovekesh Vig, TableNet: Deep Learning model for end-to-end Table detection and Tabular data extraction from Scanned Document Images, ICDAR 2020, Sydney (Main Track)
49. Kushagra Mahajan, Monika Sharma, Lovekesh Vig, Character Keypoint based homography estimation in Scanned Documents for Efficient Information Extraction, CBDAR Workshop ICDAR 2020
50. Srinidhi Hegde, Ranjitha Prasad, Ramya Hebbalaguppe, Vishwajeet Kumar: Variational Student: Learning Compact and Sparser Networks In Knowledge Distillation Framework. ICASSP 2020: 3247-3251 
51. Srinidhi Hegde, Jitender Maurya, Ramya Hebbalaguppe, Aniruddha Kalkar: SmartOverlays: A Visual Saliency Driven Label Placement for Intelligent Human-Computer Interfaces. WACV 2020: 1110-1119 
52. Varun Jain, Ramya Hebbalaguppe: AirPen: A Touchless Fingertip Based Gestural Interface for Smartphones and Head-Mounted Devices. CoRR abs/1904.06122 (2019) 
53. Varun Jain, Gaurav Garg, Ramakrishna Perla, Ramya Hebbalaguppe: GestARLite: An On-Device Pointing Finger Based Gestural Interface for Smartphones and Video See-Through Head-Mounts. CoRR abs/1904.09843 (2019) 
54. Srinidhi Hegde, Ranjitha Prasad, Ramya Hebbalaguppe, Vishwajith Kumar: Variational Student: Learning Compact and Sparser Networks in Knowledge Distillation Framework. CoRR abs/1910.12061 (2019) 
55. Varun Jain, Shivam Aggarwal, Suril Mehta, Ramya Hebbalaguppe: Synthetic Video Generation for Robust Hand Gesture Recognition in Augmented Reality Applications. CoRR abs/1911.01320 (2019)
56. Gaurav Garg, Srinidhi Hegde, Ramakrishna Perla, Varun Jain, Lovekesh Vig, Ramya Hebbalaguppe: DrawInAir: A Lightweight Gestural Interface Based on Fingertip Regression. ECCV Workshops (6) 2018: 229-240
57. Neel Rakholia, Srinidhi Hegde, Ramya Hebbalaguppe: Where to Place: A Real-Time Visual Saliency Based Label Placement for Augmented Reality Applications. ICIP 2018: 604-608
58. Jitender Maurya, Ramya Hebbalaguppe, Puneet Gupta: Real Time Hand Segmentation on Frugal Headmounted Device for Gestural Interface. ICIP 2018: 4023-4027
59. Meghal Dani, Gaurav Garg, Ramakrishna Perla, Ramya Hebbalaguppe: Mid-Air Fingertip-Based User Interaction in Mixed Reality. ISMAR Adjunct 2018: 174-178
60. Srinidhi Hegde, Gaurav Garg, Ramakrishna Perla, Ramya Hebbalaguppe: A Fingertip Gestural User Interface Without Depth Data for Mixed Reality Applications. ISMAR Adjunct 2018: 395-396
61. Ojaswi Gupta, Ramya Hebbalaguppe: FingertipCubes: an inexpensive D.I.Y wearable for 6-DoF per fingertip pose estimation using a single RGB camera. SIGGRAPH ASIA Posters 2018: 20:1-20:2
62. Kanika Kalra, Bhargav Kurma, Silpa Vadakkeeveetil Sreelatha, Manasi S. Patwardhan, Shirish Subhash Karande: Understanding Advertisements with BERT. ACL 2020: 7542-7547
63. Stephen Pilli, Manasi S. Patwardhan, Niranjan Pedanekar, Shirish Subhash Karande: Predicting Sentiments in Image Advertisements using Semantic Relations among Sentiment Labels. CVPR Workshops 2020: 1640-164
64. Mayur Patidar, Surabhi Kumari, Manasi S. Patwardhan, Shirish Subhash Karande, Puneet Agarwal, Lovekesh Vig, Gautam M. Shroff: From Monolingual to Multilingual FAQ Assistant using Multilingual Co-training. DeepLo@EMNLP-IJCNLP 2019: 115-123
65. Ameya Panse, Tushar Madheshia, Anand Sriraman, Shirish Subhash Karande: Imitation Learning on Atari using Non-Expert Human Annotations. HCOMP (WIP&Demo) 2018
66. Manasi S. Patwardhan, Richa Sharma, Abhishek Sainani, Shirish Subhash Karande, Smita Ghaisas: Moving Disambiguation of Regulations from the Cathedral to the Bazaar. HCOMP (WIP&Demo) 2018
67. Purushotam G. Radadia, Tushar Madhesia, Kanika Kalra, Anand Sriraman, Manasi S. Patwardhan, Shirish Subhash Karande: On Localizing Keywords in Continuous Speech using Mismatched Crowd. HCOMP (WIP&Demo) 2018
68. Manasi S. Patwardhan, Abhishek Sainani, Richa Sharma, Shirish Subhash Karande, Smita Ghaisas: Towards automating disambiguation of regulations: using the wisdom of crowds. ASE 2018: 850-855 
69. Ankit Sharma: MultiMBNN: Matched and Balanced Causal Inference with Neural Networks Deep Learning and Artificial Intelligence, European Symposium on Artificial Neural Networks, Computational Intelligence and Machine Learning  22.04.2020
70. Monika Sharma1, Lovekesh Vig, Monika Sharma1, Lovekesh Vig, Kushagra Mahajan, Kushagra Mahajan; Meta-DermDiagnosis: Few-Shot Skin Disease Identification using Meta-Learning; Deep Learning and Artificial Intelligence; Conference on Computer Vision and Pattern Recognition , 13.06.2020
71. Goel Nikita, Monika Sharma, Lovekesh Vig; Font-ProtoNet: Prototypical Network based Font Identification of Document Images in Low Data Regime; Deep Learning and Artificial Intelligence; Conference on Computer Vision and Pattern Recognition, 13.06.2020
72. Mouli R, Afshan Syed, Rawat Mrinal, Lovekesh Vig, Puneet A, Gautam Shroff; Information Extraction from Document Images via FCA based Template Detection and Knowledge Graph Rule Induction; Deep Learning and Artificial Intelligence; Conference on Computer Vision and Pattern Recognition, 13.06.2020

Patents - 25