Jakub_Vrabel
@VrabelJakub
Followers
181
Following
676
Statuses
93
1/12 Excited to share our new paper Input Space Mode Connectivity in Deep Neural Networks. ๐๐งต Did you know that high-dimensional input spaces in classification vision models contain simple paths between distinct (low-loss) images of the same class?
5
51
292
RT @lirarandall: Am I the only one who sees a certain irony in Sam Altman complaining anout an AI company using their data?
0
34
0
Accepted to ICLR -- see you in Singapore! ๐ฅณ
1/12 Excited to share our new paper Input Space Mode Connectivity in Deep Neural Networks. ๐๐งต Did you know that high-dimensional input spaces in classification vision models contain simple paths between distinct (low-loss) images of the same class?
0
0
3
At #NeurIPS?โฃ Come to the ๐๐๐ข๐๐๐ ๐ฐ๐จ๐ซ๐ค๐ฌ๐ก๐จ๐ฉ this Sunday at ๐ 11:05 AM to catch our ๐จ๐ซ๐๐ฅ presentation on ๐๐ง๐ฉ๐ฎ๐ญ ๐๐ฉ๐๐๐ ๐๐จ๐๐ ๐๐จ๐ง๐ง๐๐๐ญ๐ข๐ฏ๐ข๐ญ๐ฒ. I canโt make it this time, but Iโm pretty sure @DavidSKrueger will deliver an exciting talk!
1/12 Excited to share our new paper Input Space Mode Connectivity in Deep Neural Networks. ๐๐งต Did you know that high-dimensional input spaces in classification vision models contain simple paths between distinct (low-loss) images of the same class?
0
2
3
@obradam @bijbolsevika Je toto tvrdenie niecim podlozene? Pride mi to dost nepravdepodobne, kedze ipRGCs su aktivovane modrym svetlom a ziadna intenzita cerveneho svetla ich efektivne nevybudi. To je kvantova mechanika 101.
0
0
0
RT @EkdeepL: Paper alertโaccepted as a NeurIPS *Spotlight*!๐งต๐ We build on our past work relating emergence to task compositionality and anโฆ
0
91
0
@plain_simon @norabelrose Shouldnโt be even more pronounced then? The whole thing of lazy training says so. Not that surprising tbh, itโs just collective behaviorโฆ
0
0
1
RT @MBarkeshli: John Hopfield has a nice article in the annual reviews of condensed matter physics. It starts off with a discussion of whatโฆ
0
176
0
RT @SuryaGanguli: Congrats to John Hopfield and @geoffreyhinton! well deserved recognition that some important foundations of AI rest on phโฆ
0
53
0
@sams_jain @orishemur @DavidSKrueger Thank you! The paper you mention is indeed interesting and relevant, though it shows a different type of connectivity. We learned about it only after completing our work but addressed it in the Related Work section.
0
0
2
@eric_nalisnick @DavidSKrueger Not directly, we do intra-class interpolations, while Mixup does inter-class. But we've had the same question and are working on a follow-up that's more relevant to the Mixup. Hope to have better answers soon!
0
0
3
12/12 Overall, input space mode connectivity is a rich topic, raising intriguing questions and opening new directions. I truly enjoyed working on this interdisciplinary project with my amazing coauthors, @orishemur, Yaron Oz, and @DavidSKrueger. paper:
1
1
23