Correction of AI systems by linear discriminants: probabilistic foundations

DOI10.1016/J.INS.2018.07.040zbMATH Open1441.68201DBLPjournals/isci/GorbanGGMT18arXiv1811.05321OpenAlexW2884716758WikidataQ57384820 ScholiaQ57384820MaRDI QIDQ2200569FDOQ2200569

Yanyan Li

Publication date: 22 September 2020

Published in: Information Sciences (Search for Journal in Brave)

Abstract: Artificial Intelligence (AI) systems sometimes make errors and will make errors in the future, from time to time. These errors are usually unexpected, and can lead to dramatic consequences. Intensive development of AI and its practical applications makes the problem of errors more important. Total re-engineering of the systems can create new errors and is not always possible due to the resources involved. The important challenge is to develop fast methods to correct errors without damaging existing skills. We formulated the technical requirements to the 'ideal' correctors. Such correctors include binary classifiers, which separate the situations with high risk of errors from the situations where the AI systems work properly. Surprisingly, for essentially high-dimensional data such methods are possible: simple linear Fisher discriminant can separate the situations with errors from correctly solved tasks even for exponentially large samples. The paper presents the probabilistic basis for fast non-destructive correction of AI systems. A series of new stochastic separation theorems is proven. These theorems provide new instruments for fast non-iterative correction of errors of legacy AI systems. The new approaches become efficient in high-dimensions, for correction of high-dimensional systems in high-dimensional world (i.e. for processing of essentially high-dimensional data by large systems).

Full work available at URL: https://arxiv.org/abs/1811.05321

zbMATH Keywords

big dataerror correctionmeasure concentrationblessing of dimensionalitylinear discriminantnon-iterative learning

Mathematics Subject Classification ID

Probability distributions: general theory (60E05) Learning and adaptive systems in artificial intelligence (68T05) Characterization and structure theory for multivariate probability distributions; copulas (62H05) Computational aspects of data analysis and big data (68T09) Geometric probability and stochastic geometry (60D05) Inequalities; stochastic orderings (60E15) General topics in artificial intelligence (68T01)

Cites Work

• Universal approximation bounds for superpositions of a sigmoidal function
• Title not available (Why is that?)
• Title not available (Why is that?)
• On the mathematical foundations of learning
• Title not available (Why is that?)
• Observed universality of phase transitions in high-dimensional geometry, with implications for modern data analysis and signal processing
• Small ball probability estimates for log-concave measures
• Title not available (Why is that?)
• Isoperimetric and analytic inequalities for log-concave probability measures
• Concentration of measure and isoperimetric inequalities in product spaces
• Approximation of the Sphere by Polytopes having Few Vertices
• From Brunn-Minkowski to Brascamp-Lieb and to logarithmic Sobolev inequalities
• Convexity
• Title not available (Why is that?)
• Training a Support Vector Machine in the Primal
• The geometry of logconcave functions and sampling algorithms
• Interpolating thin-shell and sharp large-deviation estimates for isotropic log-concave measures
• On the shape of the convex hull of random points
• Is the \(k\)-NN classifier in high dimensions affected by the curse of dimensionality?
• Concentration property on probability spaces.
• Approximation with random bases: pro et contra
• Probabilistic lower bounds for approximation by shallow perceptron networks
• On the Geometry of Log-Concave Probability Measures with Bounded Log-Sobolev Constant
• Quasiorthogonal dimension of Euclidean spaces
• Blessing of dimensionality: mathematical foundations of the statistical physics of data
• Stochastic separation theorems
• Title not available (Why is that?)

Cited In (9)

• Fast construction of correcting ensembles for legacy artificial intelligence systems: algorithms and a case study
• One-trial correction of legacy AI systems and stochastic separation theorems
• General stochastic separation theorems with optimal bounds
• Coping with AI errors with provable guarantees
• Blessing of dimensionality at the edge and geometry of few-shot learning
• Scrutinizing XAI using linear ground-truth data with suppressor variables
• Approximation of classifiers by deep perceptron networks
• Title not available (Why is that?)
• Machine learning approach to the Floquet-Lindbladian problem

Uses Software

• FaceNet
• LFW

   Recommendations

• Fast construction of correcting ensembles for legacy artificial intelligence systems: algorithms and a case study 👍 👎
• One-trial correction of legacy AI systems and stochastic separation theorems 👍 👎
• Stochastic separation theorems 👍 👎
• General stochastic separation theorems with optimal bounds 👍 👎
• Blessing of dimensionality: mathematical foundations of the statistical physics of data 👍 👎

This page was built for publication: Correction of AI systems by linear discriminants: probabilistic foundations

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q2200569)