Online machine learning for bilateral trading problems

Αγγελοπούλου, Ειρήνη; Angelopoulou, Eirini

Άμεση μηχανική μάθηση για προβλήματα διμερών συναλλαγών

Master Thesis

Author

Αγγελοπούλου, Ειρήνη

Angelopoulou, Eirini

Date

2022

Abstract

In this work, we are studying Online Learning and its application in Online Auctions through intermediation. These days, all transactions on modern platforms are conducted exclusively through real-time online auctions. The design of such auctions and their properties have initiated a lot of research in the past years, focusing on designing the optimal auction mechanism to maximize the seller’s revenue. Specifically, throughout this study, we explore bibliographically the Online Learning and its key ideas and methods. We begin with the Full Feedback and the Expert Advice methods, followed by the Partial Feedback and the Bandit general setting. Afterwards we continue our investigation with the problem of Online Auctions and the Double Auction or Bilateral Trade setting, which we will also examine in the experimental section too. For the experiment, we explore the algorithms we presented in the section on Online Learning. Initially, we selected three Full-feedback algorithms, the Follow the Best Price algorithm, a modified Follow the Best Price with Greedy sampling, and the Exponential Weighted Average algorithm. We then continued the analysis with an ε-Greedy algorithm from the Partial-feedback category of algorithms. We compare the performance of the above algorithms in the Bilateral Trade setup, selecting four different sets for the sellers’ and the buyers’ prices. All the algorithms tested achieved small (logarithmic) regret and generally the findings verify the theoretical results. Comparing the results of the algorithms, the Follow the best price algorithm is the best performing algorithm for most of the sets while the ε-Greedy algorithm is the worst. Our modified Follow the Best Price with Greedy sample algorithm achieved the best results, actually outperforming the original Follow the Best Price algorithm, in one of the four experiments.

Postgraduate Studies Programme

Προηγμένα Συστήματα Πληροφορικής

Department

Σχολή Τεχνολογιών Πληροφορικής και Επικοινωνιών. Τμήμα Ψηφιακών Συστημάτων

Number of pages

Language

English

URI

https://dione.lib.unipi.gr/xmlui/handle/unipi/14225
http://dx.doi.org/10.26267/unipi_dione/1648

Collections

Τμήμα Ψηφιακών Συστημάτων

Show full item record

Except where otherwise noted, this item's license is described as
Αναφορά Δημιουργού-Μη Εμπορική Χρήση-Όχι Παράγωγα Έργα 3.0 Ελλάδα