Dr. Miha Bresar | Rates of convergence to stationarity with multiplicative noise: from stochastic

  Рет қаралды 19

INI Satellite Events

INI Satellite Events

Күн бұрын

Speaker: Dr Miha Bresar (University of Warwick)
Date: 8th Aug 2024 - 14:30 to 15:00
Venue: External
Title: Rates of convergence to stationarity with multiplicative noise: from stochastic reflection to denoising diffusions in generative modelling.
Event: (SSDW02) Stochastic reflection
Abstract: We characterise the rate of convergence of three classes of ergodic diffusions with multiplicative noise: (1) Reflected processes in generalised parabolic domains; (2) tempered Langevin diffusions; (3) forward processes in denoising diffusions of generative modelling.
In this talk we show that these disparate looking models share the same underlying structure, which determines their respective rates of convergence to stationarity.
(1) Varying multiplicative noise in reflecting diffusions leads to wildly different convergence rates to stationarity.
(2) Tempered Langevin diffusions with fixed invariant measure exhibit significant improvements in the rates of convergence to stationarity with the addition of unbounded multiplicative noise.
(3) We describe denoising diffusion probabilistic models (DDPMs) representing a recent advancement in generative AI, used in platforms such as ChatGPT. DDPMs are contingent on the convergence to stationarity of an underlying forward diffusion, initialised at a high dimensional multi-modal data distribution. We establish a cut-off phenomenon for the convergence of the Ornstein-Uhlenbeck forward process used in the practical applications of DDPMs and compare it with a large class of diffusions with multiplicative noise. Our results prove that in this case, unlike (1) and (2), the OU process (with additive noise) is hard to beat.
👉 Want to learn more about the talks in this stream?
🔗| Seminar Link: www.newton.ac....
🔗| Workshop Link: www.newton.ac....
-------------------
FOLLOW US
🌐| Website: www.newton.ac.uk
🎥| Main Channel: / @isaacnewtoninstitute
🐦| Twitter: / newtoninstitute
💬| Facebook: / newton.institute
📷| Instagram: / isaacnewtoninstitute
🔗| LinkedIn: / isaac-newton-institute...
SEMINAR ROOMS
🥇| INI Seminar Room 1: / @iniseminarroom1
🥈| INI Seminar Room 2: / @iniseminarroom2
🛰️| INI Satellite Events: / @inisatellite
ABOUT
The Isaac Newton Institute is a national and international visitor research institute. It runs research programmes on selected themes in mathematics and the mathematical sciences with applications over a wide range of science and technology. It attracts leading mathematical scientists from the UK and overseas to interact in research over an extended period.
👉 Learn more about us and our events here: www.newton.ac.uk

Пікірлер
Dr. Pedro Mediano | Quantifying high-order behaviours in complex systems
47:22
Bike Vs Tricycle Fast Challenge
00:43
Russo
Рет қаралды 102 МЛН
💩Поу и Поулина ☠️МОЧАТ 😖Хмурых Тварей?!
00:34
Ной Анимация
Рет қаралды 2 МЛН
Every parent is like this ❤️💚💚💜💙
00:10
Like Asiya
Рет қаралды 17 МЛН
Офицер, я всё объясню
01:00
История одного вокалиста
Рет қаралды 4,7 МЛН
AI can't cross this line and we don't know why.
24:07
Welch Labs
Рет қаралды 879 М.
Why Does Diffusion Work Better than Auto-Regression?
20:18
Algorithmic Simplicity
Рет қаралды 327 М.
The moment we stopped understanding AI [AlexNet]
17:38
Welch Labs
Рет қаралды 1,1 МЛН
Has Generative AI Already Peaked? - Computerphile
12:48
Computerphile
Рет қаралды 995 М.
Prof. Kavita Ramanan | Reflecting on Reflecting Diffusions
54:23
INI Satellite Events
Рет қаралды 7
Why Isn't Functional Programming the Norm? - Richard Feldman
46:09
Bike Vs Tricycle Fast Challenge
00:43
Russo
Рет қаралды 102 МЛН