FRINK Linked Data Fragments server

Matches in SemOpenAlex for { <https://semopenalex.org/work/W3159199716> ?p ?o ?g. }

• W3159199716 abstract "To assess generalization, machine learning scientists typically either (i) bound the generalization gap and then (after training) plug in the empirical risk to obtain a bound on the true risk; or (ii) validate empirically on holdout data. However, (i) typically yields vacuous guarantees for overparameterized models. Furthermore, (ii) shrinks the training set and its guarantee erodes with each re-use of the holdout set. In this paper, we introduce a method that leverages unlabeled data to produce generalization bounds. After augmenting our (labeled) training set with randomly labeled fresh examples, we train in the standard fashion. Whenever classifiers achieve low error on clean data and high error on noisy data, our bound provides a tight upper bound on the true risk. We prove that our bound is valid for 0-1 empirical risk minimization and with linear classifiers trained by gradient descent. Our approach is especially useful in conjunction with deep learning due to the early learning phenomenon whereby networks fit true labels before noisy labels but requires one intuitive assumption. Empirically, on canonical computer vision and NLP tasks, our bound provides non-vacuous generalization guarantees that track actual performance closely. This work provides practitioners with an option for certifying the generalization of deep nets even when unseen labeled data is unavailable and provides theoretical insights into the relationship between random label noise and generalization." @default.
• W3159199716 created "2021-05-10" @default.
• W3159199716 creator A5013136179 @default.
• W3159199716 creator A5029448258 @default.
• W3159199716 creator A5075035644 @default.
• W3159199716 creator A5079499739 @default.
• W3159199716 date "2021-05-01" @default.
• W3159199716 modified "2023-09-27" @default.
• W3159199716 title "RATT: Leveraging Unlabeled Data to Guarantee Generalization" @default.
• W3159199716 cites W1503398984 @default.
• W3159199716 cites W1798749056 @default.
• W3159199716 cites W1982723861 @default.
• W3159199716 cites W2034978228 @default.
• W3159199716 cites W2060918050 @default.
• W3159199716 cites W2064675550 @default.
• W3159199716 cites W2078394884 @default.
• W3159199716 cites W2103012681 @default.
• W3159199716 cites W2112796928 @default.
• W3159199716 cites W2139338362 @default.
• W3159199716 cites W2149298154 @default.
• W3159199716 cites W2154952480 @default.
• W3159199716 cites W2181650752 @default.
• W3159199716 cites W2194775991 @default.
• W3159199716 cites W2402144811 @default.
• W3159199716 cites W2579923771 @default.
• W3159199716 cites W2604117713 @default.
• W3159199716 cites W2618574054 @default.
• W3159199716 cites W27434444 @default.
• W3159199716 cites W2768607686 @default.
• W3159199716 cites W2809090039 @default.
• W3159199716 cites W2886067286 @default.
• W3159199716 cites W2889575872 @default.
• W3159199716 cites W2894604724 @default.
• W3159199716 cites W2904243021 @default.
• W3159199716 cites W2911742574 @default.
• W3159199716 cites W2911867426 @default.
• W3159199716 cites W2912260645 @default.
• W3159199716 cites W2915008147 @default.
• W3159199716 cites W2917123813 @default.
• W3159199716 cites W2918775908 @default.
• W3159199716 cites W2920835458 @default.
• W3159199716 cites W2952204734 @default.
• W3159199716 cites W2962698540 @default.
• W3159199716 cites W2962739339 @default.
• W3159199716 cites W2962857907 @default.
• W3159199716 cites W2962900737 @default.
• W3159199716 cites W2963236897 @default.
• W3159199716 cites W2963285844 @default.
• W3159199716 cites W2963341956 @default.
• W3159199716 cites W2963739978 @default.
• W3159199716 cites W2963794891 @default.
• W3159199716 cites W2963826371 @default.
• W3159199716 cites W2964047251 @default.
• W3159199716 cites W2964075708 @default.
• W3159199716 cites W2964098911 @default.
• W3159199716 cites W2965772785 @default.
• W3159199716 cites W2970290137 @default.
• W3159199716 cites W2970330753 @default.
• W3159199716 cites W2970971581 @default.
• W3159199716 cites W2971127900 @default.
• W3159199716 cites W2977953007 @default.
• W3159199716 cites W2996309822 @default.
• W3159199716 cites W3025991882 @default.
• W3159199716 cites W3034893234 @default.
• W3159199716 cites W3034952177 @default.
• W3159199716 cites W3035466410 @default.
• W3159199716 cites W3037144731 @default.
• W3159199716 cites W3046864714 @default.
• W3159199716 cites W3098408850 @default.
• W3159199716 cites W3102583815 @default.
• W3159199716 cites W3103317839 @default.
• W3159199716 cites W3115718026 @default.
• W3159199716 cites W3118608800 @default.
• W3159199716 cites W3132004818 @default.
• W3159199716 cites W3137695714 @default.
• W3159199716 cites W607505555 @default.
• W3159199716 hasPublicationYear "2021" @default.
• W3159199716 type Work @default.
• W3159199716 sameAs 3159199716 @default.
• W3159199716 citedByCount "2" @default.
• W3159199716 countsByYear W31591997162021 @default.
• W3159199716 crossrefType "posted-content" @default.
• W3159199716 hasAuthorship W3159199716A5013136179 @default.
• W3159199716 hasAuthorship W3159199716A5029448258 @default.
• W3159199716 hasAuthorship W3159199716A5075035644 @default.
• W3159199716 hasAuthorship W3159199716A5079499739 @default.
• W3159199716 hasConcept C107321475 @default.
• W3159199716 hasConcept C108583219 @default.
• W3159199716 hasConcept C11413529 @default.
• W3159199716 hasConcept C117765406 @default.
• W3159199716 hasConcept C119857082 @default.
• W3159199716 hasConcept C134306372 @default.
• W3159199716 hasConcept C154945302 @default.
• W3159199716 hasConcept C177148314 @default.
• W3159199716 hasConcept C177264268 @default.
• W3159199716 hasConcept C199360897 @default.
• W3159199716 hasConcept C206688291 @default.
• W3159199716 hasConcept C33923547 @default.
• W3159199716 hasConcept C41008148 @default.
• W3159199716 hasConcept C50644808 @default.

• next