• Home
  • Technology
  • Gaming
  • Entertainment
  • World & Business
  • Science
  • Sports
  • AI
HomeTechnologyGamingEntertainmentWorld & BusinessScienceSportsAI
  • HomeTechnologyGamingEntertainmentWorld & BusinessScienceSportsAI
    • Home
    • Technology
    • Gaming
    • Entertainment
    • World & Business
    • Science
    • Sports
    • AI
    AI

    X Posts Flag XM Paper Overlaps With Prior IMLE Work

    Researchers on X compare the new Explorative Modeling paper to earlier IMLE methods in algorithms, loss, and motivation.

    SZ
    AK
    YD
    30 Sources, 59d ago, first seen 59d ago

    TLDR

    Machine learning researchers posted comparisons of the Explorative Modeling paper to IMLE research. Kyle Kastner shared a review of IMLE links to the new work. You Jiacheng argued IMLE matches e2e Forward XM with amortized sampling. nrehiew_ listed matching elements across algorithm, loss, theory, and motivation in paired quotes from the paper. Susan Zhang replied that XMs extend IMLE by factoring the training loop and noted over 90 percent of results contradict IMLE theory. Jiatao Gu called IMLE inspiring. Armen Aghajanyan reported internal sweeps favoring K greater than 1 on a VLA stack.

    Combined views

    210.7K

    30 Sources, first seen 59d ago

    Combined views

    210.7K

    30 Sources, first seen 59d ago

    1.1K likes
    1.1K likes
    43 comments
    371 saves
    102 reposts

    Sentiment

    Positive——Negative

    Summary

    Not enough discussion yet.

    No sentiment analysis available yet.

    43 comments
    371 saves
    102 reposts
    Today's Rank

    —

    Not ranked yet

    Today's Rank

    —

    Not ranked yet

    Sentiment

    Positive——Negative

    Summary

    Not enough discussion yet.

    No sentiment analysis available yet.

    30 Sources

    @cloneofsimo100%.
    @thoma_guIMLE is always inspiring and insightful!
    @BlackHCRT @KL_Div: Was just as excited as everyone else to read this cool new paper, and it felt like a trip down memory lane! XM seems to be the…
    @YouJiachengIMO, IMLE ≥ e2e Forward XM. XM paper described IMLE as a special case (and seems inferior) of e2e Forward XM. But IMLE is actually e2e Forward XM with amortized candidate sampling. IMLE didn't combine best-of-K with other loss e.g. diffusion. XM did, this is the new thing.
    @PMinerviniRT @gabriberton: Reinventing the wheel and changing its name always works!
    @nrehiew_Incredible thread. Just 20 pairs of: "Both XM and IMLE ..." + "Cf. page X of the XM paper" As far as I can tell, the algorithm, the loss, the theoretical grounding and even the motivation (!) are the same?
    @kastnerkyleA nice review and overview of IMLE and relationship to recent work (along with many refs)
    @ClashlukeSample selection isn't new, with TopK-GAN, IMLE and RHO-Loss all using this lens. However, none work outside of their paper. XMs are exciting because they are the first method* where sample selection helps, is better than baseline, and scales.
    @du_yilunRT @AlexiGlad: There have been some follow-up questions on the relationship between XMs and IMLE, so I’m going to provide some clarificatio…
    @ArmenAghaOutside of this discussion, we were wondering if this would help in a mature action-policy setting. Turns out it not only doesn't help, it actively hurts. We tried hyperparameter sweeps to make it work, but no luck unfortunately.

    30 Sources

    @cloneofsimo100%.
    @thoma_guIMLE is always inspiring and insightful!
    @BlackHCRT @KL_Div: Was just as excited as everyone else to read this cool new paper, and it felt like a trip down memory lane! XM seems to be the…
    @YouJiachengIMO, IMLE ≥ e2e Forward XM. XM paper described IMLE as a special case (and seems inferior) of e2e Forward XM. But IMLE is actually e2e Forward XM with amortized candidate sampling. IMLE didn't combine best-of-K with other loss e.g. diffusion. XM did, this is the new thing.
    @PMinerviniRT @gabriberton: Reinventing the wheel and changing its name always works!
    @nrehiew_Incredible thread. Just 20 pairs of: "Both XM and IMLE ..." + "Cf. page X of the XM paper" As far as I can tell, the algorithm, the loss, the theoretical grounding and even the motivation (!) are the same?
    @kastnerkyleA nice review and overview of IMLE and relationship to recent work (along with many refs)
    @ClashlukeSample selection isn't new, with TopK-GAN, IMLE and RHO-Loss all using this lens. However, none work outside of their paper. XMs are exciting because they are the first method* where sample selection helps, is better than baseline, and scales.
    @du_yilunRT @AlexiGlad: There have been some follow-up questions on the relationship between XMs and IMLE, so I’m going to provide some clarificatio…
    @ArmenAghaOutside of this discussion, we were wondering if this would help in a mature action-policy setting. Turns out it not only doesn't help, it actively hurts. We tried hyperparameter sweeps to make it work, but no luck unfortunately.