Lightnews — Scholar-powered news

Cédric Rommel @ccrommel.bsky.social · Mar 17

Héron cendré I would say 🧐

Reposted by Cédric Rommel

Claire Vernade @claireve.bsky.social · Mar 11

I’ve put together a short list of opportunities for early career academics willing to come to Europe: www.cvernade.com/miscellaneou...

This mostly covers France and Germany for now but I’m willing to extend it. I build on @ellis.eu resources and my own knowledge of these systems.

Claire Vernade - European career opportunities

European Academic Career Opportunities in 2025

www.cvernade.com

3 26 75

Cédric Rommel @ccrommel.bsky.social · Feb 4

T-Rex scaling laws?

Randall Munroe @xkcd.com · Jan 24

T. Rex Evolution xkcd.com/3042

1

Cédric Rommel @ccrommel.bsky.social · Jan 31

Looking forward to see how it unfolds

a man in a suit is eating popcorn while sitting at a table .

ALT: a man in a suit is eating popcorn while sitting at a table .

media.tenor.com

3

Cédric Rommel @ccrommel.bsky.social · Jan 29

The way I see it is that “the market” roughly thought that scaling is all you need, so big AI actors would need to buy more and more GPUs to get better LLMs. While DeepSeek is proof that you can still get better LLMs without necessarily needing to buy more GPUs. Does that make sense?

Reposted by Cédric Rommel

Christian Wolf @chriswolfvision.bsky.social · Jan 21

Yesterday I went to Twitter and scrolled over 36h of my feed to check whether I missed something. I did not.

Even @francois.fleuret.org and @giffmana.ai seem to be less active? Guys, you should bring your skeets to this place 😀

3 2 37

Cédric Rommel @ccrommel.bsky.social · Dec 15

II had a great time at the @neurreps.bsky.social workshop today! Thanks again to the organisers of this awesome event!

Bahareh Tolooshams @btolooshams.bsky.social · Dec 14

Our first contributed talk is from @ccrommel.bsky.social on "ManiPose: Manifold-Constrained Multi-Hypothesis 3D Human Pose Estimation".

4

Cédric Rommel @ccrommel.bsky.social · Dec 14

"Does Equivariance Matter at Scale?"

Spoiler alert: Yes it does !!

I guess @johannbrehmer.bsky.social did not have to convince the @neurreps.bsky.social audience 😄

Super interesting work building scaling laws for equivariant foundation models!

Symmetry and Geometry in Neural Representations @neurreps.bsky.social · Dec 14

"Does Equivariance Matter at Scale?"

Hear from @johannbrehmer.bsky.social, a physicist turned machine learner and a research scientist at CuspAI in Amsterdam.

3

Cédric Rommel @ccrommel.bsky.social · Dec 14

Super inspiring talk by Eero Simoncelli building the bridge between inductive biases in diffusion models and how the human brain processes images!

Symmetry and Geometry in Neural Representations @neurreps.bsky.social · Dec 14

Eero Simoncelli on the geometry of the distribution of natural images

2

Cédric Rommel @ccrommel.bsky.social · Dec 13

Surprisingly clear given its high “mathematics density” :)

2

Cédric Rommel @ccrommel.bsky.social · Dec 12

It’s today @neuripsconf.bsky.social in East building poster #1500!

Whether you’re into pose estimation or just looking for a new application to test your fancy geometric deep learning method, come chat with us !!

Cédric Rommel @ccrommel.bsky.social · Dec 4

Inferring 3D human poses from video is highly ill-posed because of depth ambiguity.

Our work accepted to #NeurIPS2024, ManiPose, gets one step closer to solving this, by leveraging prior knowledge about poses topology and cool multiple-choice learning techniques.

3

Cédric Rommel @ccrommel.bsky.social · Dec 12

Very inspiring talk by Fei Fei Lee yesterday at #NeurIPS2024 on visual intelligence !

1

Cédric Rommel @ccrommel.bsky.social · Dec 11

Very interesting first invited talk at the intersection between cognitive sciences and AI by @alisongopnik.bsky.social ! 🤩

1

Cédric Rommel @ccrommel.bsky.social · Dec 4

✨Joint work with the amazing @vletzelter.bsky.social, Nermin Samet, Renaurd Marlet, Matthieu Cord, Patrick Perez and Eduardo Valle✨

Cédric Rommel @ccrommel.bsky.social · Dec 4

If you’re at #Neurips2024 next week, come meet us in poster session 3 on Thu 12 Dec 11 a.m!

Or at our oral presentation during the @neur_reps workshop on Saturday 14th!

Paper: arxiv.org/abs/2312.06386
Github: github.com/cedricrommel...

1 1

Cédric Rommel @ccrommel.bsky.social · Dec 4

We train it with the resilient winner-takes-all loss, which allows the model to optimally quantize the space without requiring many heads.

In the end, our model works as a conditional density estimator, taking the shape of a mixture of Dirac deltas.

1

Cédric Rommel @ccrommel.bsky.social · Dec 4

- Limbs length and directions are disentangled to constrain predicted poses to an estimated manifold.
- A multi-head subnetwork is used to predict different possible rotations for each joint, together with their corresponding likelihoods.
- Both are then merged into predicted poses.

1

Cédric Rommel @ccrommel.bsky.social · Dec 4

In fact, we prove the *only* way of conciliating consistency with accurate predictions is to output multiple 3D poses for each 2D input.

We hence propose ManiPose, a manifold-constrained multi-hypothesis deep network capable of better dealing with depth ambiguity.

1

Cédric Rommel @ccrommel.bsky.social · Dec 4

Previous approaches constrain poses to an estimated manifold by disentangling limbs lengths and directions. But they lag behind unconstrained models in terms of joint position error (MPJPE).

In our work, we prove this is unavoidable because of points 1 and 2.

1

Cédric Rommel @ccrommel.bsky.social · Dec 4

There are 3 main reasons to this:
1. Existing training losses and evaluation metrics (MPJPE) are blind to such inconsistencies ;
2. Many possible 3D poses can map to the same 2D input ;
3. Pose sequences cannot occupy the whole space: they lie on a smooth manifold because of limbs rigidity.

1

Cédric Rommel @ccrommel.bsky.social · Dec 4

While standard approaches directly map 2D coordinates to 3D, prior works noticed that predicted poses’ limbs could shrink and stretch along a movement.

In our work, we prove these are not isolated cases and that these methods always predict *inconsistent* 3D pose sequences.

1

Cédric Rommel @ccrommel.bsky.social · Dec 4

Many intelligent systems, like autonomous cars and smart/VR glasses, need to understand human’s movements and poses.

This can be achieved with a single camera by detecting human keypoints on a video, then lifting them into a 3D pose.

1

Cédric Rommel @ccrommel.bsky.social · Dec 4

Inferring 3D human poses from video is highly ill-posed because of depth ambiguity.

Our work accepted to #NeurIPS2024, ManiPose, gets one step closer to solving this, by leveraging prior knowledge about poses topology and cool multiple-choice learning techniques.

1 2 4

Cédric Rommel @ccrommel.bsky.social · Nov 21

Joining the 🦋 dance

a man in a suit and tie is dancing in a room

ALT: a man in a suit and tie is dancing in a room

media.tenor.com