kth.sePublications
Change search
CiteExportLink to record
Permanent link

Direct link
Cite
Citation style
  • apa
  • ieee
  • modern-language-association-8th-edition
  • vancouver
  • Other style
More styles
Language
  • de-DE
  • en-GB
  • en-US
  • fi-FI
  • nn-NO
  • nn-NB
  • sv-SE
  • Other locale
More languages
Output format
  • html
  • text
  • asciidoc
  • rtf
Topics on Large Deviations in Artificial Intelligence
KTH, School of Engineering Sciences (SCI), Mathematics (Dept.), Mathematical Statistics.ORCID iD: 0000-0001-9147-4022
2023 (English)Doctoral thesis, comprehensive summary (Other academic)
Abstract [en]

Artificial intelligence has become one of the most important fields of study during the last decade. Applications include medical sciences, autonomous vehicles, finance and everyday life. Therefore the analysis of convergence and stability of these algorithms is of utmost importance. One way of analysing the stability and speed of convergence is by the large deviations theory. In large deviations theory, a rate function characterises the exponential rate of convergence of stochastic processes. For example, by evaluating the rate function for stochastic approximation algorithms for training neural networks, faster convergence can be achieved. This thesis consists of five papers that use ideas from large deviation theory to understand and improve specific machine-learning models. 

Paper I proves that a stochastic approximation algorithm satisfies the large deviation principle with a specific rate function. This class of stochastic approximation contains many interesting learning algorithms, such as stochastic gradient descent, persistent contrastive divergence and the Wang-Landau algorithm.

Analysing the rate function from Paper I is not straightforward. In Paper II, we use tools from weak KAM theory to characterise the rate function. The rate function takes the form of a Lagrangian and can be evaluated by calculating the viscosity solution to the corresponding Hamilton-Jacobi equations. In Paper II, we also identify the projected Aubry set, a set of great importance when it comes to describing the viscosity solutions.  

Papers III, IV and V all involve Variational autoencoders (VAE), a generative deep learning model with a latent space structure. In Paper III, we develop an evaluation metric for VAEs based on large deviation theory. The idea is to measure the difference between the induced empirical measure and the prior on the latent space. This is done by training an adversarial deep neural network and proving a modified version of Sanov's theorem. 

Using the adversarial network from Paper III, we develop a stochastic interpolation algorithm for VAEs in Paper IV. The interpolation uses bridge processes and the adversarial network to construct paths that respects both the prior and generate high-quality interpolation paths.

Finally, in Paper V, a clustering algorithm is introduced. The VAE induces a probability distribution on the data space, and in this paper, we introduce an algorithm to estimate the gradient of the distribution. This leads to a stochastic approximation algorithm that gathers data in clusters. 

Abstract [sv]

Artificiell intelligens har blivit en av de viktigaste forskningsfälten de senaste åren. Användningsområden finns inom medicin forskning, självkörande fordon, finans samt vardagsbruk. Analysen av stabilitet och konvergens av dessa algoritmer har därför aldrig varit viktigare. Ett sätt att analysera dessa algoritmer är med hjälp av stora avvikelser teori. I stora avvikelser teori, en hastighets-funktion som karakteriserar den exponentiella konvergens hastigheten för stokastiska processer. Till exempel, genom att evaluera hastighets-funktionen för stokastisk approximations algoritmer för träning av neurala nätverk, snabbare konvergens kan uppnås. Den här avhandlingen består av fem artiklar som tar idéer från stora avvikelser teori för att förstå och förbättra utvalda maskininlärnings modeller. 

Artikel I bevisar att en stokastisk approximations algoritm uppfyller stora avvikelser principen med en specifik hastighest-funktion. Den här klassen av stokastisk approximation innehåller många intressanta maskininlärmnings metoder såsom, stokastisk gradient nedstigning, persistent contrastive divergence och Wang-Landau algoritmen. 

Att analysera hastighetsfunktionen från artikel I är inte enkelt. I artikel II, använder vi verktyg från svag KAM teori för att karakterisera hastighetsfunktionen. Hastighetsfunktionen är på formen av en Lagrangian och kan evalueras genom att hitta viscositetslösningar till motsvarande Hamilton-Jacobi ekvation. I Artikel II så identifierar vi den projicerade Aubry mängden, en mängd som är av stor vikt när det kommer till att beskriva viscositetslönsingar.

Artiklarna III, IV, V behandlar alla Variational autoencers (VAE), en generativ djup inlärningsmodell med latent variabel struktur. I Artikel III, utveklar vi en evaluerings metrik för VAEs baserat på stora aviklser teori. Ideen är att mäta skillnaden mellan den inducerade empiriska måttet och priori fördelningen på latenta rummet. Det åstakoms genom att träna ett adversalt nätverk och genom att bevisa en modifierad version av Sanovs sats. 

Genom att använda det adversala nätverket från artikel III vi utvecklar en stochastisk intepolations algoritm i artikel IV. Interpolations artikeln använder brygg processer och adversala nätverket för att generera interpolationer som respekterar priori fördelningen och genererar hög-kvalitativa trajektorier. 

Slutligen i artikel IV, introduceras en klustringsalgoritm. VAE inducerar en sannolikhetsförednling på data rummet, och i denna artikel, vi introducerar en algoritm för att estimera gradienten av fördelningen. Detta leder till stokastisk approximations algoritm som samlar datan i olika kluster.  

Place, publisher, year, edition, pages
KTH Royal Institute of Technology, 2023. , p. 189
Series
TRITA-SCI-FOU ; 2023:49
National Category
Probability Theory and Statistics
Research subject
Applied and Computational Mathematics, Mathematical Statistics
Identifiers
URN: urn:nbn:se:kth:diva-337383ISBN: 978-91-8040-711-3 (print)OAI: oai:DiVA.org:kth-337383DiVA, id: diva2:1801669
Public defence
2023-10-27, F3, Lindstedtsvägen 26, Stockholm, 13:00 (English)
Supervisors
Funder
Wallenberg AI, Autonomous Systems and Software Program (WASP)
Note

QC 2023-10-03

Available from: 2023-10-03 Created: 2023-10-02 Last updated: 2023-10-09Bibliographically approved
List of papers
1. A weak convergence approach to large deviations for stochastic approximations
Open this publication in new window or tab >>A weak convergence approach to large deviations for stochastic approximations
(English)Manuscript (preprint) (Other academic)
Abstract [en]

Large deviations for stochastic approximations is a well-studied field that yields convergence properties for many useful algorithms in statistics, machine learning and statistical physics. In this article, we prove, under certain assumptions, a large deviation principle for a stochastic approximation with state-dependent Markovian noise and with decreasing step size. Common algorithms that satisfy these conditions include stochastic gradient descent, persistent contrastive divergence and the Wang-Landau algorithm. The proof is based don't he weak convergence approach to the theory of large deviations and uses a representation formula to rewrite the problem into a stochastic control problem. The resulting rate function is an action potential over a local rate function that is the Fenchel-Legendre transform of a limiting Hamiltonian.

National Category
Probability Theory and Statistics
Research subject
Applied and Computational Mathematics, Mathematical Statistics
Identifiers
urn:nbn:se:kth:diva-337358 (URN)
Funder
Wallenberg AI, Autonomous Systems and Software Program (WASP)
Note

QC 20231002

Available from: 2023-10-02 Created: 2023-10-02 Last updated: 2023-10-02Bibliographically approved
2. On the projected Aubry set of the rate function associated with large deviations for stochastic approximations
Open this publication in new window or tab >>On the projected Aubry set of the rate function associated with large deviations for stochastic approximations
(English)Manuscript (preprint) (Other academic)
Abstract [en]

In this article, we look at the problem of minimizing an action potential that arises from large deviation theory for stochastic approximations. The solutions to the minimising problem satisfy, in the sense of a viscosity solution, a Hamilton-Jacobi equation. From weak KAM theory, we know that these viscosity solutions are characterised by the projected Aubryset. The main result of this paper is that, for a specific rate function corresponding to the astochastic approximation algorithm, we prove that the projected Aubry set is equal to the forward limit set to the limit ODE.

National Category
Probability Theory and Statistics
Identifiers
urn:nbn:se:kth:diva-337360 (URN)
Funder
Wallenberg AI, Autonomous Systems and Software Program (WASP)
Note

QC 20231002

Available from: 2023-10-02 Created: 2023-10-02 Last updated: 2023-10-02Bibliographically approved
3. Large Deviation Techniques for Evaluating Variational Autoencoders
Open this publication in new window or tab >>Large Deviation Techniques for Evaluating Variational Autoencoders
(English)Manuscript (preprint) (Other academic)
Abstract [en]

Evaluating generative models have become an important task in modern machine learn-ing. Despite this, most existing evaluation metrics are constricted to specific types of data orrequire a supervised setting, limiting their usefulness in the general case. In this article, wetake inspiration from large deviation theory, to propose an evaluation score for variationalautoencoders. The metric evaluates the latent space of the variational autoencoder and istherefore independent of the type of data and works in a completely unsupervised setting.Experimental results on MNIST and Fashion-MNIST prove that this new large deviationscore has a high correlation with other well know evaluation metrics.

National Category
Probability Theory and Statistics
Research subject
Applied and Computational Mathematics, Mathematical Statistics
Identifiers
urn:nbn:se:kth:diva-337357 (URN)
Funder
Wallenberg AI, Autonomous Systems and Software Program (WASP)
Note

QC 20231002

Available from: 2023-10-02 Created: 2023-10-02 Last updated: 2023-10-02Bibliographically approved
4. Particle Filter Bridge Interpolation
Open this publication in new window or tab >>Particle Filter Bridge Interpolation
(English)Manuscript (preprint) (Other academic)
National Category
Probability Theory and Statistics
Identifiers
urn:nbn:se:kth:diva-295218 (URN)
Note

QC 20210519

Available from: 2021-05-18 Created: 2021-05-18 Last updated: 2023-10-02Bibliographically approved
5. Variational Auto Encoder Gradient Clustering
Open this publication in new window or tab >>Variational Auto Encoder Gradient Clustering
(English)Manuscript (preprint) (Other academic)
National Category
Probability Theory and Statistics
Identifiers
urn:nbn:se:kth:diva-295219 (URN)
Note

QC 20210519

Available from: 2021-05-18 Created: 2021-05-18 Last updated: 2023-10-02Bibliographically approved

Open Access in DiVA

Fulltext(21161 kB)381 downloads
File information
File name FULLTEXT01.pdfFile size 21161 kBChecksum SHA-512
b4b6ddeb2f54ec079857c7c613e3d7ff4712d2b3d74bfc0b0f781d0c48bbb225947807e8c4d33d144e1d71a1e449817d62ccb62dd6d8b829e0c3d144a22c6b26
Type fulltextMimetype application/pdf
Summary(1276 kB)366 downloads
File information
File name FULLTEXT02.pdfFile size 1276 kBChecksum SHA-512
809e37546db37884a4ad90d3eb40a6f88f4aacf2b305caa1333ff9ee6b2f83dafe6db58591922c4fffb656198506999b587caa2e7ad73def90d05ccd7d478f37
Type fulltextMimetype application/pdf

Search in DiVA

By author/editor
Lindhe, Adam
By organisation
Mathematical Statistics
Probability Theory and Statistics

Search outside of DiVA

GoogleGoogle Scholar
Total: 747 downloads
The number of downloads is the sum of all downloads of full texts. It may include eg previous versions that are now no longer available

isbn
urn-nbn

Altmetric score

isbn
urn-nbn
Total: 846 hits
CiteExportLink to record
Permanent link

Direct link
Cite
Citation style
  • apa
  • ieee
  • modern-language-association-8th-edition
  • vancouver
  • Other style
More styles
Language
  • de-DE
  • en-GB
  • en-US
  • fi-FI
  • nn-NO
  • nn-NB
  • sv-SE
  • Other locale
More languages
Output format
  • html
  • text
  • asciidoc
  • rtf