Lyne Moser (EPF Lausanne)
Abstract:
A 2-category can be seen as an internal category to categories with discrete category of objects, i.e., a horizontal double category with only trivial vertical morphisms. Some aspects of 2-category theory, such as 2-limits, benefit from a passage to double categories. Going to the ∞-world, we expect to have a similar picture, which would allow one to develop aspects of (∞,2)-category theory, such as (∞,2)-limits, using double (∞,1)-categories.
A double (∞,1)-category was defined by Haugseng as a Segal object in complete Segal spaces, and then an (∞,2)-category in the form of a 2-fold complete Segal space can be interpreted as a ``horizontal\'\' double (∞,1)-category. In this talk, I will consider a slightly modified version of these double (∞,1)-categories and give a nerve construction from double categories to double (∞,1)-categories. This nerve is right Quillen and homotopically fully faithful from the category of double categories endowed with a model structure constructed in a joint work with Maru Sarazola and Paula Verdugo. By restricting along a ``homotopical\'\' horizontal embedding of 2-categories into double categories, we get a nerve from 2-categories into 2-fold complete Segal spaces, which is also right Quillen and homotopically fully faithful. I will show that these nerves are further compatible in a precise sense with the horizontal embedding of 2-categories into double categories, and this says that the ∞-setting indeed extends the strict setting.
15:00 • EPF Lausanne, ZOOM Meeting
Prof. Dr. Fernando Oliveira
A Semidefinite Programming Bound for the Average Kissing Number abstract
Abstract:
Any packing of finitely many balls (of any sizes) in R^n has a contact graph, in which the vertices are the balls and two vertices are adjacent if the balls touch. Every such contact graph of a packing has an average degree. The average kissing number of R^n is the supremum of the average degrees of contact graphs of ball packings in R^n. I will describe a semidefinite programming approach which provides the best upper bounds for the average kissing number in dimensions 3, ..., 9.(Joint work with Maria Dostert and Alexander Kolpakov.)
Prof. Dr. Samory Kpotufe (Columbia University, New York, USA)
Some Recent Insights on Transfer-Learning abstract
Abstract:
A common situation in Machine Learning is one where training data is not fully representative of a target population due to bias in the sampling mechanism or high costs in sampling the target population; in such situations, we aim to ’transfer’ relevant information from the training data (a.k.a. source data) to the target application. How much information is in the source data? How much target data should we collect if any? These are all practical questions that depend crucially on ‘how far’ the source domain is from the target. However, how to properly measure ‘distance’ between source and target domains remains largely unclear.In this talk we will argue that much of the traditional notions of ‘distance’ (e.g. KL-divergence, extensions of TV such as D_A discrepancy, density-ratios, Wasserstein distance) can yield an over-pessimistic picture of transferability. Instead, we show that some new notions of ‘relative dimension’ between source and target (which we simply term ‘transfer-exponents’) capture a continuum from easy to hard transfer. Transfer-exponents uncover a rich set of situations where transfer is possible even at fast rates, encode relative benefits of source and target samples, and have interesting implications for related problems such as multi-task or multi-source learning.In particular, in the case of multi-source learning, we will discuss (if time permits) a strong dichotomy between minimax and adaptive rates: no adaptive procedure can achieve a rate better than single source rates, although minimax (oracle) procedures can.The talk is based on earlier work with Guillaume Martinet, and ongoing work with Steve Hanneke.