diff --git a/reveal/JHU_MINDS_seminar_2024/hybrid.html b/reveal/JHU_MINDS_seminar_2024/hybrid.html index 39ca64724f..077e8a9c65 100644 --- a/reveal/JHU_MINDS_seminar_2024/hybrid.html +++ b/reveal/JHU_MINDS_seminar_2024/hybrid.html @@ -6,41 +6,25 @@

I love hybrid methods

-

-
- -
-

Infrastructure around AI is just as cool

-
-
- -
    -
  1. Autodiff
  2. -
  3. JIT compilation
  4. -
  5. Simple vectorization
  6. -
  7. GPU with XLA
  8. -
+

Reminder of why GW inference is hard

+
+
+

Multimodality

+ +

https://gracedb.ligo.org/superevents/S230601bf/view/

-
- - +
+

Small posterior (sometime)

+ +

Phys.Rev.X 9 (2019) 1, 011001

-
-

Ripple - GW model in Jax

- -
-
- -
- - \[\sim 10^6 \] - likelihood in a second, that's a whole run! -
-
+

Quick recap of MCMC

+ +

Credit: MCMC gallery

@@ -92,6 +76,41 @@

How does it work? Proposal with normalizing flow

+ +
+

Infrastructure around AI is just as cool

+
+
+ +
    +
  1. Autodiff
  2. +
  3. JIT compilation
  4. +
  5. Simple vectorization
  6. +
  7. GPU with XLA
  8. +
+
+
+ + +
+
+
+ +
+
+

Ripple - GW model in Jax

+ +
+
+ +
+ + \[\sim 10^6 \] + likelihood in a second, that's a whole run! +
+
+
+

Putting the tools together - Jim

@@ -112,20 +131,6 @@

Putting the tools together - Jim

-
-

Preliminary development - going beyond Gaussian

-
-
-

Lensing with cosmic ray

- -
-
-

GW with glitch (not yet)

- -
-
-
-