To compute the movement of the state x(t), we need to temporally integrate its velocity field ẋ(t). 🤓 The control signal steering angle stays at 0, then 0.05π, then linearly to −0.20π. The vehicle moves along circumferences. Finally, a sweep of initial velocity is performed.

Currently, writing chapter 10, «Planning and control». Physical constrains for the evolution of the state (e.g. pure rotation of the wheels) are encoded through the velocity of the state ẋ = dx(t)/dt, a function of the state x(t) and the control u(t).

Oh! The undergrad feedback came in! 🥹🥹🥹

Releasing the Energy-Book 🔋 from its first appendix's chapter, where I explain how I create my figures. 🎨 Feel free to report errors via the issues' tracker, contribute to the exercises, and show me what you can draw, via the discussion section. 🥳 github.com/Atcold/Energ...

On a summer Friday night, the first chapter sees the light. 🥹🥹🥹

Yeah, it took me 20 days to get back 🥹🥹🥹 I swear I respond to instant messages as they get through! 🥲🥲🥲 Anyhow, one more successful semester completed. 🥳🥳🥳

In this lecture from my new undergrad course, we review linear multiclass classification, leverage backprop and gradient descent to learn a linearly separable feature vector for the input, and observe the training dynamics in a 2D embedding space. 🤓 youtu.be/saskQ-EjCLQ

Training of a 2 → 100 → 2 → 5 fully connected ReLU neural net via cross-entropy minimisation. • it starts outputting small embeddings • around epoch 300 learns an identity function • takes 1700 epochs more to unwind the data manifold

Did you enjoy Alfredo Canziani's lecture as much as we did?! If so, check out his website to find more about his educational offer: atcold.github.io You can also find really cool material on Alfredo's YouTube channel! @alfcnz.bsky.social

📣 A pocos días del comienzo del Khipu 2025, nos complace anunciar que tanto las actividades del salón principal como el acto de clausura del viernes se retransmitirán en directo por este canal: khipu.ai/live/. ¡Los esperamos!

I *really* had a blast giving this improvised lecture on a topic requested on the spot without any sleep! 🤪 The audience seemed enjoying the show. 😄 To find more about my educational offer, check out my website! atcold.github.io Follow here and subscribe on YouTube! 😀

In today's episode, we review the concepts of loss ℒ(𝘄, 𝒟), per-sample loss L(𝘄, x, y), binary cross-entropy cost ℍ(y, ỹ) = y softplus(−s) + (1−y) softplus(s), ỹ = σ(𝘄ᵀ𝗳(x)). Then, we minimised the loss by choosing convenient values for our weight vector 𝘄. @nyucourant.bsky.social

Tue morning: *prepares slides* Tue class: *improv blackboard lecture* Outcome: unexpectedly great lecture. Thu morning: *prep handwritten notes* Thu class: *executes blackboard lecture* Students: 🤩🤩🤩🤩🤩🤩🤩🤩🤩 @nyucourant.bsky.social @nyudatascience.bsky.social

I think the new undergrad course is going well. At least we're having fun! 😁😁😁

If you're interested in creative coding or simulations, come read The Nature of Code with us! I'm setting up a chill book club to read the book and share projects starting in January. DM me for info!

I was reading about Conway's Game of Life, and this whole amazing book is available online! The Nature of Code - by Daniel Shiffman Thank you @shiffman.net for making it free for the web! People, please consider buying a copy of the book :) natureofcode.com/cellular-aut...

It takes me weeks to detox from a stressful work mindset and finally sleep in — without waking at dawn, driven by the haste of unfinished tasks. I struggle to turn on my laptop to do any work. When I do restart, the anxiety comes back instantaneously. Is this common? ☹️☹️☹️

Preparing an “Intro to Deep Learning” blackboard undergraduate course. Number of currently enrolled students: zero. 🥹🥹🥹 Current motivation: very weak.

You’re still arguing about tabs vs. spaces? May I present…

Entropy is one of those formulas that many of us learn, swallow whole, and even use regularly without really understanding. (E.g., where does that “log” come from? Are there other possible formulas?) Yet there's an intuitive & almost inevitable way to arrive at this expression.

Vancouver’s cannellés are delicious! 😋 Get one at Granville Island public market!

DataComp - LM

UltraPixel: Advancing Ultra-High-Resolution Image Synthesis to New Peaks

Scalable optimisation in the modular norm. Direct to the point. 😀😀😀

Scaling continuous latent variable models as probabilistic integral circuits @nolovedeeplearning.bsky.social @gengala

1/2: ✨ Google DeepMind Tokyo is growing! ✨ We're building the future of AI with Project Astra. Join my team! We're hiring a Software Engineering Manager to lead the challenge from Japan 🇯🇵 boards.greenhouse.io/deepmind/job... #GoogleDeepMind #AI #Jobs #Tokyo

Me: what’s your poster about? Presenter (P): let me tell you about the theory of constrained optimisation of convex functions. Me: no, what was the goal? P: okay, so the framework of blah and blah has the KKT blah and blah blah blah conditions and… Me: high level, plz.

Optimization Algorithm Design via Electric Circuits @StephenPBoyd1

Convolutional Differentiable Logic Gate Networks @FHKPetersen

An Eye for an Ear: Zero-shot Audio Description Leveraging an Image Captioner using Audiovisual Distribution Alignment

Outstanding presentation, finally! DynaMo: In-Domain Dynamics Pretraining for Visuo-Motor Control @jeffacce.bsky.social @lerrelpinto.com

Meta 3D AssetGen: Text-to-Mesh Generation with High-Quality Geometry, Texture, and PBR Materials

Universal Neural Functionals @AllanZhou17

BetterBench: Assessing Al Benchmarks, Uncovering Issues, and Establishing Best Practices @ankareuel.bsky.social

Hey, @neuripsconf.bsky.social, posters on the penultimate panel side are invisible. 🫥 There’s no light! This is extremely unfair for the presenter! See brightness comparison below. 👇🏻

Met the twins @roman-bushuiev.bsky.social @anton-bushuiev.bsky.social

If you're excited about #AI with #logic, check out our fully animated video on YouTube: youtu.be/FKQfMwFZvIE

Have you ever wondered how training dynamics differ between LLMs 🖋️ and Vision 👁️ models? We explore this and close the gap between VMs and LLMs in our #NeurIPS2024 paper "TrAct: Making First-layer Pre-Activations Trainable". Paper link 📜: arxiv.org/abs/2410.23970 Video link 🎥: youtu.be/ZjTAjjxbkRY 🧵

Excited to share our #NeurIPS 2024 Oral, Convolutional Differentiable Logic Gate Networks, leading to a range of inference efficiency records, including inference in only 4 nanoseconds 🏎️. We reduce model sizes by factors of 29x-61x over the SOTA. Paper: arxiv.org/abs/2411.04732

Neural Concept Binder @WolfStammer

Neural Assets: 3D-Aware Multi-Object Scene Synthesis with Image Diffusion Models @tkipf.bsky.social

TrAct: Making First-layer Pre-Activations Trainable @FHKPetersen

Generative Boba

Pink team w/ @jasonweston.bsky.social

Lorentz-Equivariant Geometric Algebra Transformers for High-Energy Physics

MoE Jetpack: From Dense Checkpoints to Adaptive Mixture of Experts for Vision Tasks

The Practitioner's Guide to Continual Multimodal Pretraining @dziadzio.bsky.social @confusezius.bsky.social @vishaalurao.bsky.social @bayesiankitten.bsky.social

Segment, Shuffle, and Stitch @Ali_Etemad1

Newton Losses: Using Curvature Information for Learning with Differentiable Algorithms @FHKPetersen