Learning in Pessiland via Inductive Inference

被引:2
|
作者
Hirahara, Shuichi [1 ]
Nanashima, Mikito [2 ]
机构
[1] Natl Inst Informat, Tokyo, Japan
[2] Tokyo Inst Technol, Tokyo, Japan
来源
2023 IEEE 64TH ANNUAL SYMPOSIUM ON FOUNDATIONS OF COMPUTER SCIENCE, FOCS | 2023年
关键词
learning; cryptography; average-case complexity; one-way function; inductive inference; FORMAL THEORY;
D O I
10.1109/FOCS57990.2023.00033
中图分类号
TP301 [理论、方法];
学科分类号
081202 ;
摘要
Pessiland is one of Impagliazzo's five possible worlds in which NP is hard on average, yet no one-way function exists. This world is considered the most pessimistic because it offers neither algorithmic nor cryptographic benefits. In this paper, we develop a unified framework for constructing strong learning algorithms under the non-existence of a one-way function, indicating a positive aspect of Pessiland. Using our framework, we improve the learning algorithm for adaptively changing distributions, which was introduced by Naor and Rothblum (ICML'06). Although the previous learner assumes the knowledge of underlying distributions, our learner is universal, i.e., does not assume any knowledge on distributions, and has better sample complexity. We also employ our framework to construct a strong agnostic learner with optimal sample complexity, which improves the previous PAC learner of Blum, Furst, Kearns, and Lipton (Crypto'93). Our learning algorithms are worst-case algorithms that run in exponential time with respect to computational depth, and as a by-product, we present the first characterization of the existence of a one-way function by the worst-case hardness of some promise problem in AM. As a corollary of our results, we establish the robustness of average-case learning, that is, the equivalence among various average-case learning tasks, such as (strong and weak) agnostic learning, learning adaptively changing distributions with respect to arbitrary unknown distributions, and weak learning with membership queries with respect to the uniform distribution. Our framework is based on the theory of Solomonoff's inductive inference and the universal extrapolation algorithm of Impagliazzo and Levin (FOCS'90). Conceptually, the framework demonstrates that Pessiland is, in fact, a wonderland for machine learning in which various learning tasks can be efficiently solved by the generic algorithm of universal extrapolation.
引用
收藏
页码:447 / 457
页数:11
相关论文
共 50 条
  • [41] Evaluation of Inductive and Transductive Inference in the context of Translation Initiation Site
    Guimaraes, Wallison W.
    Pinto, Cristiano L. N.
    Nobre, Cristiane N.
    Zarate, Luis E.
    33RD ANNUAL ACM SYMPOSIUM ON APPLIED COMPUTING, 2018, : 68 - 71
  • [42] Pedagogical Cues Influence Children's Inductive Inference and Exploratory Play
    Butler, Lucas P.
    Markman, Ellen M.
    COGNITION IN FLUX, 2010, : 1417 - 1422
  • [43] Optimal Pattern Recognition Procedures. Substantiation of Inductive Inference Procedures
    A. M. Gupal
    I. V. Sergienko
    Cybernetics and Systems Analysis, 2003, 39 (1) : 27 - 32
  • [44] Inductive inference and argumentation methods in modern intelligent decision support systems
    V. N. Vagin
    O. L. Morosin
    M. V. Fomina
    Journal of Computer and Systems Sciences International, 2016, 55 : 79 - 95
  • [45] INDUCTIVE INFERENCE OF ALGEBRAIC PROCESSES BASED ON HENNESSY-MILNER LOGIC
    TOGASHI, A
    KIMURA, S
    IEICE TRANSACTIONS ON FUNDAMENTALS OF ELECTRONICS COMMUNICATIONS AND COMPUTER SCIENCES, 1994, E77A (10) : 1594 - 1601
  • [46] Learning and approximate inference of DBNs
    Tian, FZ
    Zhang, HW
    Tian, FQ
    Lu, YC
    PROCEEDINGS OF THE 4TH WORLD CONGRESS ON INTELLIGENT CONTROL AND AUTOMATION, VOLS 1-4, 2002, : 2291 - 2296
  • [47] INDUCTIVE INFERENCE SCHEME BASED ON SPONTANEOUS FLUCTUATION OF A PROBABILISTIC NEURAL-NETWORK
    AGU, M
    YAMANAKA, K
    JAPANESE JOURNAL OF APPLIED PHYSICS PART 1-REGULAR PAPERS SHORT NOTES & REVIEW PAPERS, 1993, 32 (08): : 3666 - 3669
  • [48] Ultimate Intelligence Part II: Physical Complexity and Limits of Inductive Inference Systems
    Ozkural, Eray
    ARTIFICIAL GENERAL INTELLIGENCE (AGI 2016), 2016, 9782 : 33 - 42
  • [49] Polynomial Time Inductive Inference of TTSP Graph Languages from Positive Data
    Takami, Ryoji
    Suzuki, Yusuke
    Uchida, Tomoyuki
    Shoudai, Takayoshi
    IEICE TRANSACTIONS ON INFORMATION AND SYSTEMS, 2009, E92D (02) : 181 - 190
  • [50] Probabilistic learning and inference in schizophrenia
    Averbeck, Bruno B.
    Evans, Simon
    Chouhan, Viraj
    Bristow, Eleanor
    Shergill, Sukhwinder S.
    SCHIZOPHRENIA RESEARCH, 2011, 127 (1-3) : 115 - 122