Alladi Ramakrishnan Hall

Inference and Communication under Limited Data Access

K. R. Sahasranand

TIFR

Many a modern engineering system comprises networks of low cost edge de- vices capable of processing large amounts of data under resource constraints such as low compute-power and memory, limited communication bandwidth, and the need for low latency. Under this setting, for tasks such as inference, compression, and communication, classical al- gorithms are often inadequate and a complete paradigm shift to incorporate optimization for the practical constraints is warranted. Motivated by such applications, my research is focused on designing algorithms and lower bounds for statistical inference, compression, learning, and communication under limited data access.
In the first part of the talk, we describe an instance of statistical inference on data that is distributed across nodes and hence access to data is limited. In particular, we character- ize the communication complexity of distributed high dimensional correlation testing. Next, we make use of the idea of limited data access to envision communication under a weaker form of feedback and use it to resolve in part, a 27-year old conjecture regarding the capacity of queue channels. We conclude with a brief outline of ongoing work and future research plans.
(The talk will be in hybrid mode. Details of it are as follows.

The zoom meeting link is
zoom.us/j/99598370034
Meeting id:995 9837 0034
Passcode: 941422)