Programs & Events

programs

YINS Seminar Archives: Boaz Barak (Nov. 11, 2020)

 “Understanding generalization requires rethinking deep learning?”

Speaker: Boaz Barak, Harvard University
In classical statistical learning theory, we can place bounds on the generalization gap - the difference between the empirical performance of a learned classifier on its training set and the population performance on unseen test examples. Such bounds are hard to prove for deep learning. There is also empirical evidence that they are simply not true and deep-learning algorithms actually do have non-vanishing generalization gaps.  
 

YINS Seminar Archives: Mehryar Mohri (Mar. 10, 2021)

“Adaptation and Federated Learning”

Speaker: Mehryar Mohri, Professor of Computer Science and Mathematics, NYU

Federated learning is a prominent new learning scenario in large-scale applications, where a centralized model is trained based on data originating from a large number of clients. But, how can learning be successful with heterogenous client distributions? This talk discusses fundamental theoretical and algorithmic solutions for several variants of this problem.

YINS Seminar Archives: Dr. Matti Wilks (Feb. 24, 2021)

“The role of moral psychology in AI ethics”

Speaker: Dr. Matti Wilks, Yale University
There are many psychological factors that influence our willingness to trust AI, as well as how acceptable we think it is for AI to make decisions. In this presentation I will address the relevant literature, and extend the exploration to discuss a new project that aims to systematically investigate how our perception of AI “minds” may influence trust in AI systems.
 

“Graph Representation Learning: Recent Advances and Open Challenges”

Speaker: William L. Hamilton
Assistant Professor, School of Computer Science, McGill University

Event time: 
Wednesday, April 7, 2021 - 12:00pm to 1:15pm
Event Type: 
Weekly Seminar

“Computational Hardness of Hypothesis Testing and Quiet Plantings”

Speaker: Afonso S. Bandeira
Professor of Mathematics, ETH Zürich

Event time: 
Wednesday, March 31, 2021 - 12:00pm to 1:15pm
Event Type: 
Weekly Seminar

Yale and Tata create a more secure platform for sharing data offline

YINS
October 6, 2020

Yale researchers collaborated with Tata Consultancy Services to design a data-sharing infrastructure for environments with no internet access.

Brought together by the Yale-Tata alliance, Yale and TCS researchers introduced an off-grid network that gives users more precise control over their personal data. The network utilizes a technology called blockchain to decentralize and distribute the management of data and user information, which makes the platform more secure and reliable. The project has been going on for about two years.

“[Distributed user identities] is important because users can manage their identity and their participation in a data sharing community without having a designated node for doing identity management,” said Leandros Tassiulas, professor of electrical engineering and computer science and one of the lead researchers on the project. “This adds more robustness to the scheme.”

Currently, most off-grid networks are centralized, which means that the data is stored in a single location managed by one entity. A centralized network can be less secure than a decentralized one because it requires users to trust the administrator. It also has only one point of failure, which can be more easily exploited.

Yale researchers collaborated with Tata Consultancy Services to design a data-sharing infrastructure for environments with no internet access.

Brought together by the Yale-Tata alliance, Yale and TCS researchers introduced an off-grid network that gives users more precise control over their personal data. The network utilizes a technology called blockchain to decentralize and distribute the management of data and user information, which makes the platform more secure and reliable. The project has been going on for about two years.

“[Distributed user identities] is important because users can manage their identity and their participation in a data sharing community without having a designated node for doing identity management,” said Leandros Tassiulas, professor of electrical engineering and computer science and one of the lead researchers on the project. “This adds more robustness to the scheme.”

Currently, most off-grid networks are centralized, which means that the data is stored in a single location managed by one entity. A centralized network can be less secure than a decentralized one because it requires users to trust the administrator. It also has only one point of failure, which can be more easily exploited.

The Yale and TCS researchers instead designed a decentralized network which uses blockchain to manage data.

“Blockchain is a platform for people and entities who do not trust each other to collaborate without trusting a central entity,” said Nikolaos Papadis, a doctoral student working with Tassiulas. “It’s ideal for such a scenario where you want people and companies to collaborate and share data and services … while being sure that their data is not compromised.”

Much of the software created with blockchain is open source, which means that software libraries for implementing blockchain are freely available for anyone to use.

The team implemented several pieces of open-source software in their infrastructure: Hyperledger blockchain products to manage data and user identities, the InterPlanetary File System for file storage and an off-grid networking toolkit called MAZI for network infrastructure.

“These platforms work cohesively to secure digital identity and transfer data created in a collaborative manner,” Hanumantha Rao, Global Head in Research and Innovation at TCS Blockchain Services, wrote in an email to the News.

Once the team had a working prototype, they deployed the network onto three computers to test its performance. To evaluate the effectiveness of their infrastructure, they measured the number of transactions per second and the latency from those transactions. The team deemed their measurements acceptable for the test’s small sample size, although there was some latency in accessing data from the file system.

Although the official contract for this project has run out, more work will need to be done to deploy the team’s network on a larger scale. Tassiulas also plans to expand the platform to be able to work on mobile devices.

“I would say the product is more like a prototype,” Tassiulas said. “We have several loose ends on the platform that we are fixing in order to be fully deployable and testable.”

The next step for the platform is to tailor it to potential applications to make it more usable in different real-world situations. One situation which might call for a secure, off-grid network is small-scale electronic voting, as voting administrators might not want to expose the system to potential internet threats. This infrastructure could also be helpful for networking in rural or remote areas with limited internet connection, or right after a natural disaster that disconnects an area from the internet.

Representatives from Yale and TCS expressed that the project was beneficial for both organizations. TCS plans to apply what they learned from this project in future work, and both organizations gained insights on applications of blockchain and digital identities.

“TCS had a particular problem they wanted to solve, and Yale researchers had an interest in developing methods for solving this class of problems,” Thomas Keegan, director of the Yale Institute for Network Science and one of the managers of the Yale-Tata alliance, wrote in an email to the News.

The Yale-Tata alliance was formed in 2016.

External link: 

Pages

Subscribe to programs