CS Katha Barta | ସଂଗଣକ ବିଜ୍ଞାନ କଥା ବାର୍ତା

Hosted by Subhankar Mishra's Lab
People -> Saswat Das, Rucha Bhalchandra Joshi, Subhankar Mishra

CS Katha Barta Past years


2022

Upcoming Talks

Past Talks

  1. Dr. Makarand Tapaswi IIIT Hyderabad
    • Date: November 23, 2022, 1000 hours IST
    • Title: Fine-grained Understanding in Movies
    • Abstract

      Understanding fine-grained details in a movie (or video) requires analyzing it from various angles: what is the key action, who is performing it, with whom (or what) is it being performed, where is it happening, how or why is it occurring, and even how the characters are related. We will discuss two directions towards this goal. In the first half, we will look at MovieGraphs, a dataset for fine-grained social situation understanding and our work on joint prediction of character interactions and relationships. In the second, we will focus on situation recognition in shorter video sequences with grounding that not only improves the captions, but also helps (to some extent) resolve label ambiguity. We will end the talk with a brief discussion of open challenges in this area.

  2. Dr. Tarun Kumar Hewlett-Packard Labs, Bengaluru
    • Date: November 10, 2022,
    • Title: Algorithms for Hypergraphs: clustering and hyperedge prediction
    • Abstract

      Many real-world systems can be characterized by several components that interact in multiple ways. Such systems are observed in various domains such as brain networks in biology, railway networks in transportation systems, multiple computers connected on the internet, humans involved in multifaceted relationships, etc. These systems can be modelled by various complex graph structures such as hypergraphs, multilayer networks, knowledge graphs, multipartite graphs, etc. In this talk, I will focus on one such complex graph structure known as hypergraph. Hypergraphs are used to model systems with multiple entities interacting simultaneously. Extending traditional graph algorithms to these complex graph structures has numerous bottlenecks and demand non-trivial solutions. I will present two important problems- hypergraph clustering and hyperedge prediction. The talk will be self-contained, where I will introduce the hypergraph framework and introduce both the problems. Following this, I will discuss some of the challenges associated with the problems and how our solution addresses those. I will finish the talk by introducing some open challenges and possible directions for the future.

  3. Dr. Sumit Mukherjee Insitro (formerly at Microsoft AI for Good)
    • Date: November 4, 2022, 1130 hours IST
    • Title: Practical tools for privacy preserving synthetic data generation
  4. Dr. Pradeesha Ashok IIIT Bangalore
    • Date: October 17, 2022, 1600 hours IST
    • Title: Coloring a Dominating Set of a Graph without Conflicts
  5. Dr. Jawahar C.V. IIIT Hyderabad
    • Date: October 14, 2022, 1600 hours IST
    • Title: Emerging AIML Usecases: Challenges and Opportunities
  6. Dr. Akash Deep Oklahoma State University
    • Date: October 10, 2022, 2000 hours IST
    • Title: Event data analytics for smart and connected systems
  7. Prof. Meenakshi D'Souza IIIT Bangalore
    • Date: July 6, 2022, 1600 hours IST
    • Title: Testing of Web Applications
    • Abstract

      Web applications constitute a large class of software applications that run on the web. They can simple static websites or complex dynamic applications including web services. Web applications are developed using various technologies, with one single application involving several different languages and frameworks. This makes testing of web applications challenging and interesting. This talk will describe web applications, their features, the challenges involved in testing them and highlight a few research techniques for testing web applications.

  8. Dr. Jaya Sreevalsan Nair IIIT Bangalore
    • Date: June 29, 2022, 1600 hours IST
    • Title: Multiscale Computational Thinking in Science and Engineering
    • Abstract

      In several scientific data applications across several domains, including geoscience, and biomedicine, we encounter the need to use validation from the data itself given the absence of ground truth. Uncertainty in the data that manifests in several analytical operations must also be derived from the data itself, in several cases. In such cases, the use of multiple scales in computational methods is a powerful tool in understanding scientific and engineering data better. The definition of the scale itself changes depending on the application and the data type. Physically, scales are applicable for space and time. However, increasingly, we see scale descriptions used for image processing and other data science workflows. In this talk, we look at various examples where scales are used for a varied set of applications, to show how multiscale methods continue to be relevant.

  9. Dr. Dinesh Babu Jayagopi IIIT Bangalore
    • Date: 25 May, 2022, 1600 hours IST
    • Title: Multimodal AI for Conversations
    • Abstract

      In this talk, we will discuss three research threads. We will present the general trends, followed by our contributions. First thread involves technologies for off-line multimodal assessment and feedback in interviewing and presentation settings. While nonverbal and verbal assessment is important from the interviewer's view-point, feedback is important for the candidate. Recent trends in terms of research will be summarized, that includes both modeling and HCI developments. The second research thread concerns technologies for real-time multimodal conversational agents and applications. Multimodal analysis enables understanding user emotion and state, multimodal dialog enables making use of this information along with spoken text and subsequently generate the surface text along with the appropriate state for the virtual agent, and finally the multimodal synthesis module generates suitable nonverbal behavior and prosody for the agent reply. These systems can find applications in healthcare or education, where information can be solicited from the user and a simple task-oriented conversation can be accomplished. Innovations for quickly customizing the avatar appearance and animations are also emerging. The final research thread involves Generative Adversarial Networks (GAN) based multimodal analysis and synthesis systems to perform controlled conversational research. GANs can generate human centered images and videos, and hence enabling study of appearance and behavior manipulation, which enables controlled conversational experiments.

  10. Mr. Monik Raj Behera JP Morgan
    • Date: April 25, 2022, 1600 hours IST
    • Title: ReTAG - Real Time Anomaly Detection on Blockchain Graph
    • Abstract

      For decentralized financial services on enterprise blockchain networks like Liink by Onyx, J.P.Morgan, it is critical to ensure the messages and traffic flowing in the network as safe and anomaly free. This is to maintain a healthy network, intelligent enough to learn anomalies happening on network, in real time and to notify actors appropriately. In the solution deployed in Liink network, the model is intelligent to identify unsupervised anomalies, in real time, on streams of data and blockchain transactions. From engineering design perspective, the model is of constant space and time complexity by design, as it uses probabilistic data structure to store the model weights. Also to note, ReTAG is designed for graph data structure, where model environment can be represented as a graph. With explainable feature of the model, it is planned to be integrated with human powered feedback system, to rectify unsupervised classified labels, and adjust the weights of existing models to accommodate the human feedbacks. ReTAG has potential to identify numerous unseen anomalies, with very low rate of false positivity and can fit in multiple graph representations.

  11. Dr. Megha Khosla TU Delft
    • Date: April 21, 2022, 1600 hours IST
    • Title: Privacy in Graph Neural Networks
    • Abstract

      Graph Neural Networks (GNNs), which generalize traditional deep neural networks on graph data, have achieved state-of-the-art performance on several graph analytical tasks. With their growing popularity in various applications including sensitive applications like health it is imperative to study the privacy aspects of these models. In this talk I will discuss the vulnerability of GNNs to privacy leakage and also present our framework on releasing GNNs under differential privacy guarantees.

  12. Dr. Charu Sharma IIIT Hyderabad
    • Date: April 11, 2022, 1600 hours IST
    • Title: Geometric Deep Learning for Graphs and Point Clouds
    • Abstract

      Recent years have seen a surge in cutting-edge research on geometric machine learning for complex structured data such as 3D point clouds, graphs, and manifolds. While traditional ML has mainly focused on data in the Euclidean domain, in comparison, geometric ML aims to generalize to non-Euclidean domains. However, working with the non-Euclidean domain is accompanied by its own set of research challenges, namely, varying sizes of vertex neighborhoods, as opposed to images with regular grids and fixed neighborhood sizes. Additionally, there are no standard distance metrics between graphs. In this talk, we will focus on geometric representation learning for 3D point clouds and briefly on graphs.

  13. Mr. Abhinav Agshikar Amazon
    • Date: April 7, 2022. 1030 hours IST
    • Title: Multi-Tenancy in a SaaS environment
    • Abstract

      Tenant isolation is fundamental to the design and development of Software as a Service (SaaS) systems. It enables SaaS providers to reassure customers that—even in a multitenant environment—their resources cannot be accessed by other tenants. This talk will cover architectural patterns that solve tenancy and associated problems such as Noisy Neighbor. We will also deep dive into advanced topics and solutions that AWS implements to achieve Tenant Isolation.

    • Resources:
      1. AWS SaaS Whitepaper
      2. AWS Shuffle Sharing Lab
      3. Shuffle Sharing Blog
  14. Prof. I. A. Palani IIT Indore
    • Date: March 28, 2022, 1600 hours IST
    • Title: Shape memory alloy actuated soft robotic systems
  15. Mr. Sumeet Gupta Akamai Technologies
    • Date: March 16, 2022, 1100 hours IST
    • Title: Say Hello to SASE: Secure Access Service Edge
  16. Dr. Ankit Gangwal IIIT Hyderabad
    • Date: February 19, 2022, 1100 hours IST
    • Title: Introduction to Layer-two Blockchains