Jamolbek Mattiev

Urgench State University, Uzbekistan

Title: Associative classification model based on clustering

Abstract: The size of collected data is increasing and the number of
rules generated on those datasets is getting bigger. Producing compact
and accurate models is being the most important task of data mining. In
this talk, we present a new associative classifier that utilizes
agglomerative hierarchical clustering. Experimental evaluations show
that proposed method achieves significantly better results than
classical rule learning algorithms in terms of rules on bigger datasets
while maintaining classification accuracy on those datasets.

Michael Murray

University of California, Los Angeles

Title:  An Introduction to Benign Overfitting

Abstract: Conventional machine learning wisdom suggests that the generalization error of a complex model will typically be worse versus a simpler model when both are trained to interpolate data. Indeed, the bias-variance trade-off implies that although choosing a complex model is advantageous in terms of approximation error, it comes at the price of an increased risk of overfitting. The traditional solution to managing this trade-off is to use some form of regularization, allowing the optimizer to select a predictor from a rich class of functions while at the same time encouraging it to choose one that is in some sense simple. However, in recent years it has been observed that many models, including deep neural networks, trained with minimal if any form of explicit regularization, can almost perfectly interpolate noisy data with nominal cost to their generalization performance. This phenomenon is referred to as benign or tempered overfitting and there is now great interest in characterizing it mathematically. In this talk I’ll give an introduction and motivation for the topic, describe some of the key results derived so far and and highlight open questions.

Michael Porter

University of Virginia

Title: Modeling contagion, excitation, and social influence with Hawkes point processes

Abstract: Many social and physical processes (e.g., crime, conflict, social media activity, financial markets, new product adoption, social network communication, earthquakes, neural spiking, disease spread) produce event point patterns that exhibit clustering. Hawkes, or self-exciting point process, models are a popular choice for modeling the clustering patterns which can be driven by both exogenous influences and endogenous forces like contagion/self-excitement. These models stipulate that each event can be triggered by past events creating a branching structure that produces the endogenous clustering. The contagion effects are modeled by a shot-noise term which aggregates the influence of past events to temporarily increase the event rate following each event. This talk will introduce the hawkes process and illustrate some extensions and uses of hawkes models in three areas: modeling contagion in terrorist attacks, forecasting crime hotspots, and identifying social influence in Yelp restaurant reviews.

Rishi Sonthalia

University of California, Los Angeles

Title: From Classical Regression to the Modern Regime: Surprises for Linear Least Squares Problems

Abstract: Linear regression is a problem that has been extensively studied. However, modern machine learning has brought to light many new and exciting phenomena due to overparameterization. In this talk, I briefly introduce the new phenomena observed in recent years. Then, building on this, I present recent theory work on linear denoising. Despite the importance of denoising in modern machine learning and ample empirical work on supervised denoising, its theoretical understanding is still relatively scarce. One concern about studying supervised denoising is that one might not always have noiseless training data from the test distribution. It is more reasonable to have access to noiseless training data from a different dataset than the test dataset. Motivated by this, we study supervised denoising and noisy-input regression under distribution shift. We add three considerations to increase the applicability of our theoretical insights to real-life data and modern machine learning. First, we assume that our data matrices are low-rank. Second, we drop independence assumptions on our data. Third, the rise in computational power and dimensionality of data have made it essential to study non-classical learning regimes. Thus, we work in the non-classical proportional regime, where data dimension $d$ and number of samples N grow as d/N = c + o(1). For this setting, we derive general test error expressions for both denoising and noisy-input regression and study when overfitting the noise is benign, tempered, or catastrophic. We show that the test error exhibits double descent under general distribution shifts, providing insights for data augmentation and the role of noise as an implicit regularizer. We also perform experiments using real-life data, matching the theoretical predictions with under 1% MSE error for low-rank data.

Angelica Torres

Max Planck Institute, Germany

Title: Algebraic Geometry meets Structure from Motion

Abstract:  The Structure from Motion (SfM) pipeline aims to create a 3D model of a scene using two-dimensional images as input. The process has four main stages: Feature detection, matching, camera pose, and triangulation. In the first step, features such as points and lines are detected in the images, then they are matched to features appearing in other images. After the matching stage, the actual images are forgotten and the data that remains are tuples of points or lines that are believed to come from the same world object. This is geometric data, hence the toolbox coming from Algebraic Geometry can be used to estimate the camera positions and to triangulate the objects in the scene. In this talk I will introduce the SfM pipeline, and present some of the algebraic varieties that arise when the pinhole camera model is assumed. During the talk we will highlight how some properties of the varieties translate into properties of data and how it can affect the image reconstruction process.