Event-chain algorithms: taming randomness in Monte Carlo methods through irreversibility, factorization and lifting
Duration: 39 mins 44 secs
Share this media item:
Embed this media item:
Embed this media item:
About this item
| Description: |
Manon, M
Tuesday 18th July 2017 - 12:10 to 12:50 |
|---|
| Created: | 2017-07-19 12:12 |
|---|---|
| Collection: | Scalable inference; statistical, algorithmic, computational aspects |
| Publisher: | Isaac Newton Institute |
| Copyright: | Manon, M |
| Language: | eng (English) |
| Abstract: | I will first present the irreversible and rejection-free Monte Carlo methods recently developed in Physics under the name Event-Chain. They have proven to produce clear acceleration over standard Monte Carlo methods, thanks to the reduction of their random-walk behavior. Their irreversible nature relies on three key ingredients: the factorized filter, the generalized lifting framework and the infinitesimal moves. Then, I will focus on the new Forward Event-Chain version that allows to reduce the randomization needed for ergodicity, leading to a striking speed-up. Finally, I will explain how the factorized filter may be the key to subsampling in Monte Carlo methods. |
|---|---|
Available Formats
| Format | Quality | Bitrate | Size | |||
|---|---|---|---|---|---|---|
| MPEG-4 Video | 640x360 | 1.94 Mbits/sec | 578.70 MB | View | Download | |
| WebM | 640x360 | 438.98 kbits/sec | 127.81 MB | View | Download | |
| iPod Video | 480x270 | 522.23 kbits/sec | 151.98 MB | View | Download | |
| MP3 | 44100 Hz | 249.75 kbits/sec | 72.77 MB | Listen | Download | |
| Auto * | (Allows browser to choose a format it supports) | |||||