Agenda:
2:30-3 pm, Dr. Sivaram Ambikasaran- Talk Title: Fast algorithms for large scale kernel machines
Abstract:
Kernel machines are frequently encountered in many machine learning problems. However, one of the downsides of using kernel machines is that they are slow to compute especially on large datasets. In this talk, we will look at a few ways to make kernel machines scalable for large datasets.

3-3:30 pm: Prof. Nandan Sudarsanam, Talk Title: Quantifying the Maximum Possible Improvement in 2^k Experiments
Abstract:
This research formulates, and numerically quantifies the optimal response that can be discovered in a design space characterized by main effects, and two-way and three-way interactions. In an experimental design setup, this can be conceptualized as the response of the best treatment combination of a 2^k full factorial design. Using Gaussian and Uniform priors for the strength of main effects and interaction effects, this study enables a practitioner to make estimates of the maximum possible improvement that is possible through design space exploration. For basic designs up to two factors, we construct the full distribution of the optimal treatment. Whereas, for values of k ≥ 3, we analytically formulate two indicators of a greedy heuristic of the expected value of the optimal treatment. We present results for these formulations up to k = 7 factors and validate these through simulations. Finally, we also present an illustrative case study of the power loss in disengaged wet clutches, which confirms our findings and serves as an implementation guide for practitioners.

3:30 pm- 4 pm, Dr. Lakshmi Narasimhan, Talk Title: Resource Allocation for Generalized Gaussian Multiple Access Channels
Abstract:
In this talk, we will consider multiple access channels (MAC) with generalized gaussian (GG) noise. Several wireless systems such as WiFi, cellular and satellite networks can be described using the GG-MAC model. For this model, we shall discuss asymptotic and finite-regime resource allocation strategies that maximize the total average data-rate. It will be shown that the greedy power allocation policy can be optimal even for the finite-user scenario. Finally, a multiarm bandits formulation of the problem will be discussed.

4:00-4:30 pm, Prof. Manikandan Narayanan, Talk Title: From multi-modal data to insights into multi-cellular life: the role of multi-layer/causal network methods
Abstract:
As a research group, we are interested in dissecting the molecular (e.g., gene-gene, gene-protein, gene-metabolite, etc.) interactions underlying multicellular life, and investigating how these interactions rewire in a complex disease. We use the language of networks, specifically multi-layer networks, to represent and study such molecular interactions. A multi-layer network is a collection of graphs, with graphs in different layers typically defined over the same set of nodes, but linked via distinct sets of edges (both within the layer and to other layers). In this talk, I will give an overview of the methods we've developed to infer such multi-layer and related causal networks from observational data, and to analyze them to derive nodes with high within-layer or global centrality in the network. The predictions and related insights from our methods, a subset of which pertaining to a neurodegenerative disease are being tested experimentally, are promising and encourage taking a computational lens to probe complex biological systems.