<div dir="ltr"><div>Please join us for the fourth <span class="gmail-il">theory lunch</span> of the quarter today!</div><div dir="auto"><br></div><div dir="auto"><div dir="auto"><b>Time: </b>Wednesday, April 16, 2025, 12pm-1pm in JCL 390</div><div dir="auto"> <br><b>Speaker:</b> Andrzej Kaczmarczyk</div><div dir="auto"><br><b>Title: </b>Learning Real-Life Approval Elections</div><div dir="auto"><br><b>Abstract</b>: We study how to learn an approval election, i.e., an election in which each voter selects which candidates they approve. Specifically, we focus on the independent approval model (IAM), where each candidate has its own approval probability and is approved independently of the other ones. We propose algorithms for learning IAMs and their mixtures from data, using either maximum likelihood estimation or Bayesian learning. We then apply these algorithms to a large set of real-life elections. In particular, we find that single-component models are rarely sufficient to capture the complexity of real-life data, whereas their mixtures perform well.</div></div></div><br><div class="gmail_quote gmail_quote_container"><div dir="ltr" class="gmail_attr">On Tue, Apr 8, 2025 at 1:05 PM Alec Sun <<a href="mailto:alecsun@uchicago.edu">alecsun@uchicago.edu</a>> wrote:<br></div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex"><div dir="ltr"><div dir="ltr"><div>Please join us for the third theory lunch of the quarter tomorrow!</div><div dir="auto"><br></div><div dir="auto"><div dir="auto"><b>Time: </b>Wednesday, April 9, 2025, 12pm-1pm in JCL 390</div><div dir="auto"> <br><b>Speaker: </b><span style="font-family:Arial;font-size:10pt">Subhodh Kotekal</span></div><div dir="auto"><br><b>Title: </b><span style="font-family:Arial;font-size:10pt">Variance estimation in compound decision theory under boundedness</span></div><div dir="auto"><br><b>Abstract</b>: <span style="font-family:Arial;font-size:10pt">The normal means model is often studied under the assumption of a known variance. However, ignorance of the variance is a frequent issue in applications and basic theoretical questions still remain open in this setting. This article establishes that the sharp minimax rate of variance estimation in square error is $(\log\log n/\log n)^2$ under arguably the most mild assumption imposed for identifiability: bounded means. The rate-optimal estimator proposed in this article achieves the optimal rate by estimating $O\left(\log n/\log\log n\right)$ cumulants and leveraging a variational representation of the noise variance in terms of the cumulants of the data distribution. The minimax lower bound involves a moment matching construction.</span></div></div></div><br><div class="gmail_quote"><div dir="ltr" class="gmail_attr">On Tue, Apr 1, 2025 at 1:38 PM Alec Sun <<a href="mailto:alecsun@uchicago.edu" target="_blank">alecsun@uchicago.edu</a>> wrote:<br></div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex"><div dir="auto"><div>Please join us for the second theory lunch of the quarter tomorrow!</div><div dir="auto"><br></div><div dir="auto"><div dir="auto"><b>Time: </b>Wednesday, April 2, 2025, 12pm-1pm in JCL 390</div><div dir="auto"> <br><b>Speaker: </b>Dravy Sharma</div><div dir="auto"><br><b>Title: </b>Provable tuning of deep learning model hyperparameters</div><div dir="auto"><br><b>Abstract</b>: Modern machine learning algorithms, especially deep learning-based techniques, typically involve careful hyperparameter tuning to achieve the best performance. Despite the surge of intense interest in practical techniques like Bayesian optimization and random search-based approaches to automating this laborious and compute-intensive task, the fundamental learning-theoretic complexity of tuning hyperparameters for deep neural networks is poorly understood. Inspired by this glaring gap, we initiate the formal study of hyperparameter tuning complexity in deep learning under a powerful data-driven paradigm. A major difficulty is that the utility function as a function of the hyperparameter is very volatile and furthermore, it is given implicitly by an optimization problem over the model parameters. To tackle this challenge, we employ subtle concepts from differential/algebraic geometry and constrained optimization to show that the learning-theoretic complexity of the corresponding family of utility functions is bounded. We instantiate our results and provide sample complexity bounds for concrete applications—tuning a hyperparameter that interpolates neural activation functions and setting the kernel parameter in graph neural networks. The talk is based on joint work with Nina Balcan and Anh Nguyen.</div><br><div class="gmail_quote" dir="auto"><div dir="ltr" class="gmail_attr">On Tue, Mar 25, 2025, 4:52 PM Alec Sun <<a href="mailto:alecsun@uchicago.edu" rel="noreferrer noreferrer" target="_blank">alecsun@uchicago.edu</a>> wrote:<br></div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex"><div dir="ltr">Please join us for the first theory lunch of the quarter tomorrow!<div> </div><div><b>Time: </b>Wednesday, March 26, 2025, 12pm-1pm in JCL 390</div><div> <br><b>Speaker: </b>Olga Medrano</div><div><br><b>Title: </b>Short Visit to Regularity Lemmas</div><div><br><b>Abstract: </b>In this expository talk, we will overview Szémerédi’s regularity lemma, a result from extremal graph theory with different applications. We will then mention some considerations about this result, including the existence of irregular pairs in the partitions obtained, as well as the large size of those partitions. Time permitting, we very briefly reflect on how the proof of Szémerédi's regularity lemma is not algorithmic and mention a few lines of work that were focused on finding algorithms to output regular partitions. In the second half of this talk, we will describe versions of this lemma over certain classes of graphs. In particular, we state both the Ultra-strong regularity lemma, which works for the class of graphs of bounded VC dimension, and the Stable regularity lemma, which works for the class of k-edge stable graphs (namely, those not containing a bi-induced bipartite graph). We conclude by acknowledging that algorithmic questions for both results remain open.</div></div>
</blockquote></div></div></div>
</blockquote></div></div>
</blockquote></div>