Scott H. Hawley
@drscotthawley.bsky.social
Mostly: ML for music production workflows.
Professor of Physics & Senior Data Fellow at Belmont University, Nashville TN
Head of Research for Hyperstate Music AI.
Teacher of audio engineers, Opinions my own.
Explainer blog: https://drscotthawley.github.io
Professor of Physics & Senior Data Fellow at Belmont University, Nashville TN
Head of Research for Hyperstate Music AI.
Teacher of audio engineers, Opinions my own.
Explainer blog: https://drscotthawley.github.io
The submission deadline has been extended to *December 15* for the joint JAES/DAFx Special Issue, "New Frontiers in Digital Audio Effects"! aes2.org/wp-content/u...
November 5, 2025 at 2:46 PM
The submission deadline has been extended to *December 15* for the joint JAES/DAFx Special Issue, "New Frontiers in Digital Audio Effects"! aes2.org/wp-content/u...
Wasn't Universe 25 the one where all the mice died? @github.com Education, are you saying we're going to eat each other?
October 30, 2025 at 1:50 PM
Wasn't Universe 25 the one where all the mice died? @github.com Education, are you saying we're going to eat each other?
haha, This was all "honor system". However, the reason we made the models unconditional was because this week's tutorial is on Guidance! So, this week we used the classifier to impose classifier guidance on the models. I'll release a new blog post on this very soon.
October 24, 2025 at 1:08 AM
haha, This was all "honor system". However, the reason we made the models unconditional was because this week's tutorial is on Guidance! So, this week we used the classifier to impose classifier guidance on the models. I'll release a new blog post on this very soon.
Been working on a tutorial on Guidance for (latent) flow-based generative models, and happened upon a great metaphor today while taking a little "me" time. :-)
October 16, 2025 at 6:26 PM
Been working on a tutorial on Guidance for (latent) flow-based generative models, and happened upon a great metaphor today while taking a little "me" time. :-)
At this rate, Trons #10 through Infinity will all be released in 2042! ;-)
October 16, 2025 at 5:18 AM
At this rate, Trons #10 through Infinity will all be released in 2042! ;-)
Claude has a "weekly limit" now? I'm already a Pro subscriber! :-/
October 15, 2025 at 6:52 PM
Claude has a "weekly limit" now? I'm already a Pro subscriber! :-/
Running a Leaderboard contest in my Deep Learning & AI Ethics class: combine previous lessons into a full latent flow-matching generative model (for MNIST digits). Deadline's tomorrow for students, will later release materials so others can try. 2025-dlaie-leaderboard.streamlit.app
October 11, 2025 at 2:05 PM
Running a Leaderboard contest in my Deep Learning & AI Ethics class: combine previous lessons into a full latent flow-matching generative model (for MNIST digits). Deadline's tomorrow for students, will later release materials so others can try. 2025-dlaie-leaderboard.streamlit.app
Here's (part of) an answer for Exercise 5: namely a fully convolutional VAE with residual connections, for which the latent space is a "wee image":
gist.github.com/drscotthawle...
gist.github.com/drscotthawle...
October 3, 2025 at 4:28 AM
Here's (part of) an answer for Exercise 5: namely a fully convolutional VAE with residual connections, for which the latent space is a "wee image":
gist.github.com/drscotthawle...
gist.github.com/drscotthawle...
New lesson! "An Arsenal of AutoEncoders (AE, VAE, Conditional VAE, VQ-VAE)" Colab link:
github.com/drscotthawle...
github.com/drscotthawle...
September 25, 2025 at 9:46 PM
New lesson! "An Arsenal of AutoEncoders (AE, VAE, Conditional VAE, VQ-VAE)" Colab link:
github.com/drscotthawle...
github.com/drscotthawle...
I wonder if all the job postings requiring "publication in NeurIPS, ICLR, ICML, or similar conferences" might have anything to do with the massive glut of submissions to these venues.
September 23, 2025 at 4:05 PM
I wonder if all the job postings requiring "publication in NeurIPS, ICLR, ICML, or similar conferences" might have anything to do with the massive glut of submissions to these venues.
PREACH IT CLAUDE!
September 9, 2025 at 4:04 AM
PREACH IT CLAUDE!
Oh one more. From a talk in March 2022: hedges.belmont.edu/Taming_Chaos...
July 31, 2025 at 6:12 AM
Oh one more. From a talk in March 2022: hedges.belmont.edu/Taming_Chaos...
Here's one I generated "back in the day", though I think this was actually pre-SD Midjourney... so maybe my timeline's a bit off. ;-) It was a busy summer. www.instagram.com/p/CbdgHzPFTSj/
July 31, 2025 at 6:01 AM
Here's one I generated "back in the day", though I think this was actually pre-SD Midjourney... so maybe my timeline's a bit off. ;-) It was a busy summer. www.instagram.com/p/CbdgHzPFTSj/
One example, here's a post from @andy.baio.net around that time with an Afremov vibe... waxy.org/2022/08/expl...
July 31, 2025 at 5:55 AM
One example, here's a post from @andy.baio.net around that time with an Afremov vibe... waxy.org/2022/08/expl...
I learned of Afremov during the summer-2022 heyday of Stable Diffusion when freelance/hobbyist researchers shared lists of artist-prompt-style pairings. (Sadly most of the researchers' names got removed due to misdirected backlash) Cool opportunity to support artists' fam afremov.com/deal-of-the-...
July 31, 2025 at 5:55 AM
I learned of Afremov during the summer-2022 heyday of Stable Diffusion when freelance/hobbyist researchers shared lists of artist-prompt-style pairings. (Sadly most of the researchers' names got removed due to misdirected backlash) Cool opportunity to support artists' fam afremov.com/deal-of-the-...
"You Are Absolutely Correct! Yes-Men ARE Worse Than Useless as Advisors!" ...is my proposed talk for an upcoming Baylor symposium on "Technology and the Human Person in the Age of AI" (ifl.web.baylor.edu/conferences-...) No apologies. ;-)
July 28, 2025 at 6:53 PM
"You Are Absolutely Correct! Yes-Men ARE Worse Than Useless as Advisors!" ...is my proposed talk for an upcoming Baylor symposium on "Technology and the Human Person in the Age of AI" (ifl.web.baylor.edu/conferences-...) No apologies. ;-)
Mission accomplished.
July 19, 2025 at 12:25 AM
Mission accomplished.
FLUX.1 Kontext helped me create an appropriate YouTube thumbnail for this...
July 16, 2025 at 3:42 PM
FLUX.1 Kontext helped me create an appropriate YouTube thumbnail for this...
Adding it to my list of teaching resources! I don't know JAX/FLAX at all, so this is interesting.
Super-minor typo: looks like Colab doesn't render \mathcal{N}: (at least in my browser, Brave.)
Super-minor typo: looks like Colab doesn't render \mathcal{N}: (at least in my browser, Brave.)
June 27, 2025 at 5:29 PM
Adding it to my list of teaching resources! I don't know JAX/FLAX at all, so this is interesting.
Super-minor typo: looks like Colab doesn't render \mathcal{N}: (at least in my browser, Brave.)
Super-minor typo: looks like Colab doesn't render \mathcal{N}: (at least in my browser, Brave.)
Monday afternoon at #IJCNN I'll give a tutorial on the full dev pipeline for conditional latent flow-matching generative models --we'll talk a LOT about autoencoders! -- plus "inference tricks" e.g. inpainting. This follows Jesse Lai's (Sony AI)'s tutorial on diffusion models. Come add discussion!
June 27, 2025 at 5:26 PM
Monday afternoon at #IJCNN I'll give a tutorial on the full dev pipeline for conditional latent flow-matching generative models --we'll talk a LOT about autoencoders! -- plus "inference tricks" e.g. inpainting. This follows Jesse Lai's (Sony AI)'s tutorial on diffusion models. Come add discussion!
New to me.
VQ (commit) loss is the first to go NaN, then MSE, then cross-entropy.
commitment_weight = 0.6
decay = 0.95,
threshold_ema_dead_code = 2,
rotation_trick = True,
orthogonal_reg_weight=0.2,
2 codebooks, 16 vectors each, 2 dims per vector.
3 of the level-1 vectors look almost unused:
VQ (commit) loss is the first to go NaN, then MSE, then cross-entropy.
commitment_weight = 0.6
decay = 0.95,
threshold_ema_dead_code = 2,
rotation_trick = True,
orthogonal_reg_weight=0.2,
2 codebooks, 16 vectors each, 2 dims per vector.
3 of the level-1 vectors look almost unused:
June 26, 2025 at 7:03 PM
New to me.
VQ (commit) loss is the first to go NaN, then MSE, then cross-entropy.
commitment_weight = 0.6
decay = 0.95,
threshold_ema_dead_code = 2,
rotation_trick = True,
orthogonal_reg_weight=0.2,
2 codebooks, 16 vectors each, 2 dims per vector.
3 of the level-1 vectors look almost unused:
VQ (commit) loss is the first to go NaN, then MSE, then cross-entropy.
commitment_weight = 0.6
decay = 0.95,
threshold_ema_dead_code = 2,
rotation_trick = True,
orthogonal_reg_weight=0.2,
2 codebooks, 16 vectors each, 2 dims per vector.
3 of the level-1 vectors look almost unused:
Been visualizing the heck out of my latent flow-matching models lately.
Re. "What does Vector Quantization do?": If you just look at the locations of vectors, they may look like a gaussian-shaped "blob". But within that blob can be very different probabilities, vs. the smooth shape of a gaussian.
Re. "What does Vector Quantization do?": If you just look at the locations of vectors, they may look like a gaussian-shaped "blob". But within that blob can be very different probabilities, vs. the smooth shape of a gaussian.
June 13, 2025 at 5:45 PM
Been visualizing the heck out of my latent flow-matching models lately.
Re. "What does Vector Quantization do?": If you just look at the locations of vectors, they may look like a gaussian-shaped "blob". But within that blob can be very different probabilities, vs. the smooth shape of a gaussian.
Re. "What does Vector Quantization do?": If you just look at the locations of vectors, they may look like a gaussian-shaped "blob". But within that blob can be very different probabilities, vs. the smooth shape of a gaussian.
Call for Papers! AES+DAFx Special Issue, "New Frontiers in Audio Effects" - note that despite the title the list of topics is quite broad & even includes MIR.
Deadline: December 1, 2025.
Planned publication: Summer 2026.
Link to CFP: users.spa.aalto.fi/vpv/JAES_V73...
Deadline: December 1, 2025.
Planned publication: Summer 2026.
Link to CFP: users.spa.aalto.fi/vpv/JAES_V73...
June 10, 2025 at 4:02 PM
Call for Papers! AES+DAFx Special Issue, "New Frontiers in Audio Effects" - note that despite the title the list of topics is quite broad & even includes MIR.
Deadline: December 1, 2025.
Planned publication: Summer 2026.
Link to CFP: users.spa.aalto.fi/vpv/JAES_V73...
Deadline: December 1, 2025.
Planned publication: Summer 2026.
Link to CFP: users.spa.aalto.fi/vpv/JAES_V73...
If/when you release the training code, I'd love to try it on my dataset! Which is arguably out-of-distribution for common image tasks. (re. arxiv.org/abs/2407.01499)
June 8, 2025 at 7:56 PM
If/when you release the training code, I'd love to try it on my dataset! Which is arguably out-of-distribution for common image tasks. (re. arxiv.org/abs/2407.01499)