Browsing by Author "Steifer, Tomasz"

Now showing 1 - 2 of 2
  • Thumbnail Image
    Item
    Find a witness or shatter: the landscape of computable PAC learning
    (2023) Delle Rose, Valentino; Kozachinskiy, Alexander; Rojas González, Luis Cristóbal; Steifer, Tomasz
    This paper contributes to the study of CPAC learnability -- a computable version of PAC learning -- by solving three open questions from recent papers. Firstly, we prove that every improperly CPAC learnable class is contained in a class which is properly CPAC learnable with polynomial sample complexity. This confirms a conjecture by Agarwal et al (COLT 2021). Secondly, we show that there exists a decidable class of hypothesis which is properly CPAC learnable, but only with uncomputably fast growing sample complexity. This solves a question from Sterkenburg (COLT 2022). Finally, we construct a decidable class of finite Littlestone dimension which is not improperly CPAC learnable, strengthening a recent result of Sterkenburg (2022) and answering a question posed by Hasrati and Ben-David (ALT 2023). Together with previous work, our results provide a complete landscape for the learnability problem in the CPAC setting.
  • Thumbnail Image
    Item
    No Agreement Without Loss: Learning and Social Choice in Peer Review
    (2023) Barcelo Baeza, Pablo; Duarte, Mauricio; Rojas González, Luis Cristóbal; Steifer, Tomasz
    In peer review systems, reviewers are often asked toevaluate various features of submissions, such as technical qualityor novelty. A score is given to each of the predefined features andbased on these the reviewer has to provide an overall quantitativerecommendation. It may be assumed that each reviewer has her ownmapping from the set of features to a recommendation, and thatdifferent reviewers have different mappings in mind. This introducesan element of arbitrariness known as commensuration bias. In thispaper we discuss a framework, introduced by Noothigattu, Shah andProcaccia, and then applied by the organizers of the AAAI 2022conference. Noothigattu, Shah and Procaccia proposed to aggregatereviewer’s mapping by minimizing certain loss functions, and studiedaxiomatic properties of this approach, in the sense of social choicetheory. We challenge several of the results and assumptions used intheir work and report a number of negative results. On the one hand,we study a trade-off between some of the axioms proposed and theability of the method to properly capture agreements of the majorityof reviewers. On the other hand, we show that dropping a certainunrealistic assumption has dramatic effects, including causing themethod to be discontinuous.