On Asymptotic and Finite-Time Optimality of Bayesian Predictors
Authors: Daniil Ryabko
JMLR 2019 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Theoretical | The problem is that of sequential probability forecasting for finite-valued time series. ... It is shown that the minimax asymptotic average loss which may be positive is always attainable, and it is attained by a Bayesian predictor whose prior is discrete and concentrated on C. Moreover, the finite-time loss of the Bayesian predictor is also optimal up to an additive log n term (where n is the time step). This upper bound is complemented by a lower bound that goes to infinity but may do so arbitrarily slow. ... Two results are presented to support the suggested approach to the general problem of sequential prediction. The first result shows that, no matter how big a model class C is, if the distribution generating the data belongs to C then the optimal minimax asymptotic average performance, even if positive, is always achievable, and it can be attained by a Bayesian combination of countably many distributions in C. ... Theorem 1 For every set C of probability measures and for every predictor ρ there is a discrete Bayesian predictor ν such that for every µ C we have Ln(µ, ν) Ln(µ, ρ) + 8 log n + O(log log n). |
| Researcher Affiliation | Academia | Daniil Ryabko EMAIL. The author is listed with a personal email domain 'EMAIL' and no institutional affiliation. However, given the publication venue (Journal of Machine Learning Research) and the highly theoretical nature of the content, the work is primarily academic. Without a corporate affiliation or a mix, it is classified as academia. |
| Pseudocode | No | The paper describes theoretical concepts, theorems, and proofs using mathematical notation and natural language. There are no explicitly labeled pseudocode blocks, algorithms, or structured steps formatted like code. |
| Open Source Code | No | The paper does not contain any statements about releasing source code, links to code repositories, or mentions of code in supplementary materials. |
| Open Datasets | No | The paper discusses theoretical concepts related to data sequences (e.g., 'piece-wise i.i.d. sequence', 'Bernoulli i.i.d. measure') as examples or subjects of theoretical analysis, but it does not use or provide access to any specific, publicly available datasets for empirical evaluation. |
| Dataset Splits | No | The paper is theoretical and does not conduct experiments on datasets, therefore no dataset split information (training, validation, test) is provided. |
| Hardware Specification | No | The paper is theoretical and does not describe any experimental setup that would require hardware. Therefore, no hardware specifications (like CPU, GPU models, or cloud resources) are mentioned. |
| Software Dependencies | No | The paper is theoretical and does not mention any software implementations or dependencies with specific version numbers. |
| Experiment Setup | No | The paper focuses on theoretical results, theorems, and proofs, and does not include any experimental setup details such as hyperparameters, training configurations, or system-level settings. |