Tutorials

A Step-by-Step Derivation of a Metadata Private Voice Communication System over Untrusted Infrastructure

Amr El Abbadi (University of California, Santa Barbara)

Metadata from voice calls, such as the knowledge of who is communicating with whom, contains rich information about people’s lives. Indeed, it is a prime target for powerful adversaries such as nation states. Existing systems that hide voice call metadata either require trusted intermediaries in the network or scale to only tens of users. In this tutorial, we will develop, in a step-by step pedagogical manner, Addra, the first system for voice communication that hides metadata over fully untrusted infrastructure and scales to tens of thousands of users. At a high level, Addra follows a template in which callers and callees deposit and retrieve messages from private mailboxes hosted at an untrusted server. However, Addra improves message latency in this architecture, which is a key performance metric for voice calls. First, it enables a caller to push a message to a callee in two hops, using a new way of assigning mailboxes to users that resembles how a post office assigns PO boxes to its customers. Second, it innovates on the underlying cryptographic machinery and constructs a new private information retrieval scheme, FastPIR, which reduces the time to process oblivious access requests for mailboxes.

This tutorial does not assume background cryptography knowledge and helps the audience understand and appreciate the challenges faced by practical designers of privacy preserving systems when using sophisticated cryptographic techniques, including homomorphic encryption. The derivation is pedagogical: motivating from a systems point of view the various optimizations needed to develop the final efficient privacy preserving system. Furthermore, the derived private information scheme has many applications in large scale privacy preserving data management systems.

Based on:

Ishtiyaque Ahmad, Yuntian Yang, Divyakant Agrawal, Amr El Abbadi, Trinabh Gupta:
Addra: Metadata-private voice communication over fully untrusted infrastructure. OSDI 2021
https://www.usenix.org/conference/osdi21/presentation/ahmad

Time: 1.5 Hours

Speaker’s Bio: Amr El Abbadi is a Professor of Computer Science at the University of California, Santa Barbara. He received his B. Eng. from Alexandria University, Egypt, and his Ph.D. from Cornell University. His research interests are in the fields of fault-tolerant distributed systems and databases, focusing recently on Cloud data management and blockchain based systems. Prof. El Abbadi is an ACM Fellow, AAAS Fellow, and IEEE Fellow. He was Chair of the Computer Science Department at UCSB from 2007 to 2011. He has served as a journal editor for several database journals, including, The VLDB Journal, IEEE Transactions on Computers and The Computer Journal. He has been Program Chair for multiple database and distributed systems conferences. He currently serves on the executive committee of the IEEE Technical Committee on Data Engineering (TCDE) and was a board member of the VLDB Endowment from 2002 to 2008. In 2007, Prof. El Abbadi received the UCSB Senate Outstanding Mentorship Award for his excellence in mentoring graduate students. In 2013, his student, Sudipto Das received the SIGMOD Jim Gray Doctoral Dissertation Award. Prof. El Abbadi is also a co-recipient of the Test of Time Award at EDBT/ICDT 2015. He has published over 300 articles in databases and distributed systems and has supervised over 35 PhD students.

Automated Trading using API

Jagdish Chakole (Symbiosis Institute of Technology, Nagpur) and Manish Kurhekar (VNIT Nagpur)

Automation has played a significant role in many domains, and stock market trading is no exception. Even retail traders can automate his/her trading strategy using API. A computer program doing trading is known as Algorithmic Trading. It eliminates inefficiency due to human emotions. Trading logic that decides when to buy and sell a stock is generally termed as a trading strategy. The availability of large trading data has made it possible to automate the generation of dynamic trading strategies. Speed and accuracy are very vital aspects of profitable stock market trading. Algorithmic trading is far superior to manual trading in terms of speed and accuracy. In this tutorial, we demonstrate the automation of predefined trading strategies using Python API. We will also demonstrate the generation of trading strategies using reinforcement learning, Deep learning, and various other domains in computer science. Validating the performance of any predefined trading strategy on historical data plays a significant role in its live performance and it is known as backtesting. Backtesting is the key feature of Algorithmic Trading. We will demonstrate backtesting of the trading strategies using a computer program on the Indian and American stock market data."

AI & Law: Formative Developments, State-of-the-Art Approaches, Challenges & Opportunities

Jack G. Conrad (Thomson Reuters), Shirsha Ray Chaudhuri (Thomson Reuters), Shounak Paul (IIT Kharagpur) and Saptarshi Ghosh (IIT Kharagpur)

Artificial Intelligence (AI), Machine Learning (ML), and Natural Language Processing (NLP) are transforming the way legal professionals and law firms approach their work. The significant potential for the application of AI to Law, for instance, by creating computational solutions for legal tasks, has intrigued researchers for decades. This appeal has only been amplified with the advent of Deep Learning (DL). In particular, research in AI & Law can be extremely beneficial in countries like India with an overburdened legal system.

In this tutorial, we will give an overview of the various aspects of applying AI to legal textual data. We will start with a history of AI & Law, and then discuss the current state of AI & Law research including the techniques that have produced the biggest impact. We will also take a deep dive into the software processes required to implement and sustain such AI solutions.

AI for Immersive Metaverse Experience

Alpana Dubey (Accenture), Nitish Bhardwaj (Accenture), Abhinav Upadhyay (Accenture) and Roshni Ramnani (Accenture)

Metaverse has received a huge attention in recent times with several Big Techs having invested in this concept. While Metaverse may mean different things to different people, at the core, the Metaverse is a web of interactive 3D immersive spaces that enables a user to move beyond ‘browsing’ to ‘inhabiting’ in a persistent, shared experience that spans the spectrum of our real world to the fully virtual and in between. The evolution that Metaverse brings can be seen along three dimensions: 1) shift towards spatial experiences: which includes 2D, 3D, augmented, virtual, and mixed reality immersive experiences, 2) shared co-presence: where users experience a persistent shared space with a sense of co-presence with others, and 3) trusted identities and transactions to address challenges of fake identities, products, and transactions as present in today’s internet. This evolution opens an enormous opportunity to rethink the digital experiences future applications would offer to the people. AI would be the core engine behind making these experiences richer, immersive, and engaging. The role of AI, in the Metaverse, is broad. In this tutorial, we will focus on two areas where AI will play a major role in shaping up the form and function of the Metaverse by: 1) bringing more realism in Metaverse with high fidelity immersive content generated through AI techniques and 2) enhancing user interactions by bringing more intelligence in the interaction modes.

Causal Inference and Causal Machine Learning with Practical Applications

Somedip Karmakar (Walmart Global Tech India), Soumojit Guha Majumder (Walmart Global Tech India) and Dhiraj Gangaraju (Walmart Global Tech India)

One of the most important research areas in Machine Learning is to build prescriptive models. This requires understanding and measurement of the causal impact of any proposed treatment, followed by designing optimal strategy based on such causal estimation. Traditional impact measurement frameworks like A-B testing & Randomized Control Trials have certain limitations in terms of feasibility, time constraint, and unknown confounders. Observational Causal Inference techniques can achieve similar and better results in terms of measuring impact of new proposed changes to any systems. Our models serve a wide variety of users who may respond very differently to any changes, so such heterogeneous behaviour can be well captured through Causal Machine Learning models which helps in developing better prescriptive recommendations and implementation strategies. The tutorial will cover techniques of observational causal inference like Propensity and Covariate matching, Regression Discontinuity Design, Difference-in-difference, Causal ML techniques of conditional average treatment effect estimation, using wide variety of algorithms like meta-learners, direct uplift estimation, tree-based algorithms, Double Debiased ML models, Deep Learning based methods. It will also cover model validation and visualization for Causal ML models, along with implementation of such models in industry case studies.

DL4HC: Deep Learning for Healthcare

Raghavendra Bhat (Intel), Sandya Mannarswamy (Independent Researcher) and Shreyas N C (Intel)

Over the last few years, Machine Learning (ML), and particularly Deep Learning (DL), has made great strides and has been successfully deployed in many real-world applications such as healthcare, customer care, finance, autonomous driving etc. One of the recent applications of deep learning has been in healthcare. ML/DL techniques have been applied to healthcare in medical imaging, clinical decision making, electronic healthcare records processing etc. In particular, Federated Learning (FL), an important sub-area of DL has been considered as a critical part of advancing ML driven healthcare systems in a data siloed healthcare ecosystem. FL addresses the critical need to train ML/DL algorithms without needing to move data to a central location. This tutorial is intended to provide a high-level overview of DL4HC at a major data science/ML research venue such as CODS-COMAD to focus research attention in this emerging area. This tutorial provides a detailed overview of DL/FL techniques in healthcare. While there has been considerable progress in this area, several challenges remain, especially in applying probabilistic techniques to healthcare which often requires deterministic correctness, interpretability, and verifiability. We discuss the various challenges involved in applying DL/FL techniques to healthcare and outline some of the future research directions. We believe that having this tutorial at CODS-COMAD would be an important step in facilitating increased research focus from data science community and greater collaboration with mainstream healthcare community on this topic.

Promoting sustainable water usage and management with water data, AI and policy

Bihu Suchetana (IIT Roorkee), Biplav Srivastava (University of South Carolina), Hari Prabhat Gupta (IIT-BHU) and Manabendra Saharia (IIT Delhi)

Abstract: This tutorial will introduce the audience to the potential and challenges of using AI techniques on water data from India and the US to promote sustainable usage. The insights can guide not only near term end-user decisions, like what water to discharge in a water body or where to draw in water for cultivation, but also long-term water policy decisions like where to build a sewage treatment plant. The team consists of researchers and practitioners with expertise in water, sensing, data management and AI techniques who have worked on similar problems across the globe.

Semantic Automation for Data Discovery

Rajmohan C (IBM Research), Ritwik Chaudhuri (IBM Research), Balaji Ganesan (IBM Research), Anupam Sanghi (IBM Research), Arvind Agarwal (IBM Research) and Sameep Mehta (IBM Research)

Data discovery is a multi-dimensional field encompassing information extraction, information retrieval, exploratory data analysis, visualization and recommendations among other things. Data Marketplaces are platforms where users discover and shop for data products. These products themselves are produced by modern data stacks governed by frameworks like Data Fabric. Knowledge Graphs and semantic technologies already form a core part of Data Fabric and hence could be leveraged for data discovery. In this tutorial, we'll present state of the art semantic technologies that enable automation of various tasks in data discovery. In particular, we'll focus on data enrichment, datasets search and recommendations, and explorations within a dataset.

Understanding meta-learning for fast adaptation

Kavya R. Saxena (IIT Kanpur) and Vipul Arora (IIT Kanpur)

Deep learning methods provide excellent performance when sufficiently large annotated data is available. But their performance degrades when the models are applied to different target domains with very different data distributions as compared to that of the source domain. There are many techniques that cater to this problem, such as transfer learning, domain adaptation, few-shot learning, and meta-learning.

In this tutorial, we discuss these techniques with a methodical analysis of meta-learning. Meta-learning works in a manner similar to how humans take in a handful of examples and learn new skills very quickly and efficiently. Therefore, meta-learning focuses on speeding the few-shot learning method. Generally, data-level and parameter-level approaches are considered when solving a few-shot learning problem. Meta-learning is one of the parameter-level approaches. In this tutorial, we will also cover various approaches to meta-learning.

The tutorial is targeted toward those who wish to gain comprehensive insight and practical experience in applying meta-learning for fast adaptation. We show applications of these techniques through Jupyter notebooks, demonstrating how core concepts can be translated into empirical work. For those who are new to this area, we will provide a hands-on session and online pedagogical guide that can serve as the basis for fostering future research. We hope to convey a comprehensive understanding of recent advances and current state-of-the-art approaches to those who have prior knowledge of the areas. We would expect some basic experience in the execution of Jupyter notebooks and some basic knowledge of concepts related to machine learning as a prerequisite for participation.