Universal Online Learning with Unbounded Losses: Memory Is All You Need

被引:0
作者
Blanchard, Moise [1 ]
Cosson, Romain [1 ]
Hanneke, Steve [2 ]
机构
[1] MIT, Cambridge, MA 02139 USA
[2] Purdue Univ, W Lafayette, IN 47907 USA
来源
INTERNATIONAL CONFERENCE ON ALGORITHMIC LEARNING THEORY, VOL 167 | 2022年 / 167卷
关键词
online learning; universal consistency; stochastic processes; measurable partitions; statistical learning theory; Borel measure; SEQUENTIAL PREDICTION;
D O I
暂无
中图分类号
TP18 [人工智能理论];
学科分类号
081104 ; 0812 ; 0835 ; 1405 ;
摘要
We resolve an open problem of [13] on the subject of universally consistent online learning with non-i.i.d. processes and unbounded losses. The notion of an optimistically universal learning rule was defined by Hanneke [13] in an effort to study learning theory under minimal assumptions. A given learning rule is said to be optimistically universal if it achieves a low long-run average loss whenever the data generating process makes this goal achievable by some learning rule. [13] posed as an open problem whether, for every unbounded loss, the family of processes admitting universal learning are precisely those having a finite number of distinct values almost surely. In this paper, we completely resolve this problem, showing that this is indeed the case. As a consequence, this also offers a dramatically simpler formulation of an optimistically universal learning rule for any unbounded loss: namely, the simple memorization rule already suffices. Our proof relies on constructing random measurable partitions of the instance space and could be of independent interest for solving other open questions [14]. We extend the results to the non-realizable setting thereby providing an optimistically universal Bayes consistent learning rule.
引用
收藏
页数:21
相关论文
共 28 条
  • [1] Ben-David S., 2009, COLT
  • [2] Nonparametric sequential prediction of time series
    Biau, Gerard
    Bleakley, Kevin
    Gyorfi, Laszlo
    Ottucsak, Gyoergy
    [J]. JOURNAL OF NONPARAMETRIC STATISTICS, 2010, 22 (03) : 297 - 317
  • [3] A Theory of Universal Learning
    Bousquet, Olivier
    Hanneke, Steve
    Moran, Shay
    van Handel, Ramon
    Yehudayoff, Amir
    [J]. STOC '21: PROCEEDINGS OF THE 53RD ANNUAL ACM SIGACT SYMPOSIUM ON THEORY OF COMPUTING, 2021, : 532 - 541
  • [4] Chapelle O., 2006, SEMISUPERVISED LEARN
  • [5] Cohen Dan Tsir, 2022, Metric-valued regression
  • [6] NEAREST NEIGHBOR PATTERN CLASSIFICATION
    COVER, TM
    HART, PE
    [J]. IEEE TRANSACTIONS ON INFORMATION THEORY, 1967, 13 (01) : 21 - +
  • [7] Devroye L., 2013, PROBABILISTIC THEORY, V31
  • [8] Evans SN, 2023, Arxiv, DOI arXiv:2012.12859
  • [9] Gyofi L aszlo., 2002, Modeling uncertainty, P225
  • [10] A simple randomized algorithm for sequential prediction of ergodic time series
    Györfi, L
    Lugosi, G
    Morvai, G
    [J]. IEEE TRANSACTIONS ON INFORMATION THEORY, 1999, 45 (07) : 2642 - 2650