Video wird geladen...

Video konnte nicht geladen werden

Zur Startseite

Deep Learning architectures usually aren't trained to perform search at test time, leading to sample inefficiency + poor generalization. Latent Program Network (LPN) builds in test-time adaption by learning a latent space that can be searched. Clem Bonnet matt

ndea's profile picture

Ndea

10,789 subscribers

30,803 Aufrufe • vor 1 Jahr •via X (Twitter)

BildungWissenschaft & Technologie
LIVE

Anya Rossi• Live Now

Private livecam show

0 Kommentare

Keine Kommentare verfügbar

Kommentare vom Original-Post werden hier angezeigt

Ähnliche Videos

DeCAF won the #ICML Test of Time Award 2024! Big congrats to trevordarrell (my PhD advisor at MIT), and Jeff Donahue. 🎉 You may not heard of DeCAF, but it is everywhere! DeCAF stands for Deep Convolutional Activation Features. Published ten years ago, the DeCAF paper is a groundbreaking work that shows the activation features of the last few layers of a deep network contain useful features that can be "repurposed" for or "transferred to" many other tasks, not just the original task the network was trained for. I created this exercise to show where we can see DeCAF's influence in some of the most well-known architectures: AlexNet, ViT, U-Net, CLIP, and Latent Diffusion, to prove that DeCAF's "Test of Time Award" is well-deserved! Let's give a round of applause to DeCAF, the unsung hero of computer vision.
0:10

DeCAF won the #ICML Test of Time Award 2024! Big congrats to trevordarrell (my PhD advisor at MIT), and Jeff Donahue. 🎉 You may not heard of DeCAF, but it is everywhere! DeCAF stands for Deep Convolutional Activation Features. Published ten years ago, the DeCAF paper is a groundbreaking work that shows the activation features of the last few layers of a deep network contain useful features that can be "repurposed" for or "transferred to" many other tasks, not just the original task the network was trained for. I created this exercise to show where we can see DeCAF's influence in some of the most well-known architectures: AlexNet, ViT, U-Net, CLIP, and Latent Diffusion, to prove that DeCAF's "Test of Time Award" is well-deserved! Let's give a round of applause to DeCAF, the unsung hero of computer vision.

Tom Yeh

21,420 Aufrufe • vor 1 Jahr

Been dying for this to arrive! 🤯 Limited edition. You can only find it in the latent space.
0:10

Been dying for this to arrive! 🤯 Limited edition. You can only find it in the latent space.

Javi Lopez ⛩️

87,978 Aufrufe • vor 1 Jahr

ICML 2026: Latent Reasoning in TRMs is Secretly a Policy Improvement Operator Why does recursive reasoning, especially latent reasoning, actually work? The theory is still young, and even mechanistic explanations are limited. We close part of this gap by showing that latent reasoning is secretly doing policy improvement. Each recursion pushes the model steadily toward the target. Based on this view, we propose an algorithm that boosts learning and inference efficiency by up to 18x.
0:12

ICML 2026: Latent Reasoning in TRMs is Secretly a Policy Improvement Operator Why does recursive reasoning, especially latent reasoning, actually work? The theory is still young, and even mechanistic explanations are limited. We close part of this gap by showing that latent reasoning is secretly doing policy improvement. Each recursion pushes the model steadily toward the target. Based on this view, we propose an algorithm that boosts learning and inference efficiency by up to 18x.

Arip

24,553 Aufrufe • vor 16 Tagen

The latent space of earlier generative models like GANS can linearly encode concepts of the data. What if the data was model weights? We present weights2weights, a subspace in diffusion weights that behaves as an interpretable latent space over customized diffusion models.
0:16

The latent space of earlier generative models like GANS can linearly encode concepts of the data. What if the data was model weights? We present weights2weights, a subspace in diffusion weights that behaves as an interpretable latent space over customized diffusion models.

Amil Dravid

94,276 Aufrufe • vor 2 Jahren

Learn how to combine a deep learning model for pose estimation to perform a 3D reconstruction using two cameras 📸
0:13

Learn how to combine a deep learning model for pose estimation to perform a 3D reconstruction using two cameras 📸

MATLAB

15,699 Aufrufe • vor 1 Jahr

Is it possible to adapt a neural network on the fly at the test time to cope with distribution shifts? RNA does precisely that by creating a closed-loop feedback system. We will present it on Wed afternoon at #ICCV2025. 1/n
0:13

Is it possible to adapt a neural network on the fly at the test time to cope with distribution shifts? RNA does precisely that by creating a closed-loop feedback system. We will present it on Wed afternoon at #ICCV2025. 1/n

Amir Zamir

21,685 Aufrufe • vor 2 Jahren

1. Machine Learning Specialization Break into AI with this 3-course program by Andrew Ng. What you’ll learn: → Build machine learning models → Train neural networks → Deep reinforcement learning → Unsupervised learning techniques 🔗
0:10

1. Machine Learning Specialization Break into AI with this 3-course program by Andrew Ng. What you’ll learn: → Build machine learning models → Train neural networks → Deep reinforcement learning → Unsupervised learning techniques 🔗

Amit

55,462 Aufrufe • vor 6 Monaten

Apple Learning Coach is a free professional learning program that trains instructional coaches, digital learning specialists, and other coaching educators to help teachers effectively use Apple technology in the classroom. Apply today.
0:18

Apple Learning Coach is a free professional learning program that trains instructional coaches, digital learning specialists, and other coaching educators to help teachers effectively use Apple technology in the classroom. Apply today.

Apple Education

25,869 Aufrufe • vor 3 Jahren

u js trained on test bro, it's not that deep
0:16

u js trained on test bro, it's not that deep

llm_enjoyer

197,332 Aufrufe • vor 1 Monat

POV: you agree to test a non slip bonnet
0:14

POV: you agree to test a non slip bonnet

MNJ

133,087 Aufrufe • vor 1 Jahr

CS projects in 2015: Snake game in Python that took you 2 weeks to make. CS projects in 2030: Deep Learning Snake Game Neural Network.
0:11

CS projects in 2015: Snake game in Python that took you 2 weeks to make. CS projects in 2030: Deep Learning Snake Game Neural Network.

ₕₐₘₚₜₒₙ

132,224 Aufrufe • vor 2 Jahren

How can we use test-time compute for spatial understanding? 🤔 In InterPose, we propose to repeatedly sample generative video models to help two-view pose estimation and reconstruction, by leveraging the video models' keyframe interpolation abilities. A 🧵... (1/8)
0:17

How can we use test-time compute for spatial understanding? 🤔 In InterPose, we propose to repeatedly sample generative video models to help two-view pose estimation and reconstruction, by leveraging the video models' keyframe interpolation abilities. A 🧵... (1/8)

Ricardo Martin-Brualla

21,685 Aufrufe • vor 1 Jahr

We’re all looking at the latent space, but ⁦gray ❄️🫧⁩ is inside of it.
0:10

We’re all looking at the latent space, but ⁦gray ❄️🫧⁩ is inside of it.

Max

11,388 Aufrufe • vor 1 Jahr

One week left to apply! ⏰ Apple Learning Coach is a free professional learning program that trains instructional coaches, digital learning specialists, and other educators to help teachers get more out of Apple technology.
0:18

One week left to apply! ⏰ Apple Learning Coach is a free professional learning program that trains instructional coaches, digital learning specialists, and other educators to help teachers get more out of Apple technology.

Apple Education

27,059 Aufrufe • vor 3 Jahren

1. Machine Learning Specialization. Break Into AI with this 3-course program by Andrew Ng. What You’ll learn: → Build ML models → Train neural networks → Deep reinforcement learning → Unsupervised learning techniques 🔗
0:13

1. Machine Learning Specialization. Break Into AI with this 3-course program by Andrew Ng. What You’ll learn: → Build ML models → Train neural networks → Deep reinforcement learning → Unsupervised learning techniques 🔗

Abhishek

364,323 Aufrufe • vor 1 Jahr

At Nodepay, we’ve built a living ecosystem fueled by unused bandwidth, real-time data retrieval, user-owned AI innovation, & rewards that flow back to the contributors. Picture a global AI network constantly learning and evolving—all powered by our community. How it works: 🧵
0:13

At Nodepay, we’ve built a living ecosystem fueled by unused bandwidth, real-time data retrieval, user-owned AI innovation, & rewards that flow back to the contributors. Picture a global AI network constantly learning and evolving—all powered by our community. How it works: 🧵

Nodepay

965,030 Aufrufe • vor 1 Jahr

The term "continual learning" has become overloaded if you see it as an ML problem. One classic thread is about memorization: regularization-based continual learning methods, such as EWC, MAS, and SI, estimate which parameters mattered for previous tasks and resist changing them too much. One modern thread is about adaptation: test-time training and inference-time learning methods, such as TTT, adapt part of the model on the incoming test stream before making predictions. These are sometimes discussed as separate threads. But in modern scalable architectures, I think they are better seen as complementary constraints: a model that learns quickly at test time also benefits from a mechanism for deciding what not to forget. In our #ECCV2026 paper, we study this in large-scale 4D reconstruction: how to build fast spatial memory that can adapt over long observation streams while reducing collapse and forgetting. Instead of using fully plastic test-time updates, we stabilize fast-weight adaptation with an elastic prior that balances adaptation and memory. Key ideas: - Elastic Test-Time Training: Fisher-weighted consolidation for fast-weight updates - EMA anchor weights that provide a moving reference for stability - Chunk-by-chunk inference for long 3D/4D observation streams We show that this scales across large 3D/4D pretraining settings, including both LRM-style and LVSM-style models, and improves reconstruction across benchmarks including Stereo4D, NVIDIA, and DL3DV-140. We release model checkpoints across different design choices: resolution, post-training curriculum, and whether the model uses an explicit 4DGS intermediate representation. - Homepage: - Paper: - Code: - Models: This work is co-led with Xueyang Yu, contributed by Haoyu Zhen Yuncong Yang, and advised by Michigan SLED Lab Chuang Gan.
0:14

The term "continual learning" has become overloaded if you see it as an ML problem. One classic thread is about memorization: regularization-based continual learning methods, such as EWC, MAS, and SI, estimate which parameters mattered for previous tasks and resist changing them too much. One modern thread is about adaptation: test-time training and inference-time learning methods, such as TTT, adapt part of the model on the incoming test stream before making predictions. These are sometimes discussed as separate threads. But in modern scalable architectures, I think they are better seen as complementary constraints: a model that learns quickly at test time also benefits from a mechanism for deciding what not to forget. In our #ECCV2026 paper, we study this in large-scale 4D reconstruction: how to build fast spatial memory that can adapt over long observation streams while reducing collapse and forgetting. Instead of using fully plastic test-time updates, we stabilize fast-weight adaptation with an elastic prior that balances adaptation and memory. Key ideas: - Elastic Test-Time Training: Fisher-weighted consolidation for fast-weight updates - EMA anchor weights that provide a moving reference for stability - Chunk-by-chunk inference for long 3D/4D observation streams We show that this scales across large 3D/4D pretraining settings, including both LRM-style and LVSM-style models, and improves reconstruction across benchmarks including Stereo4D, NVIDIA, and DL3DV-140. We release model checkpoints across different design choices: resolution, post-training curriculum, and whether the model uses an explicit 4DGS intermediate representation. - Homepage: - Paper: - Code: - Models: This work is co-led with Xueyang Yu, contributed by Haoyu Zhen Yuncong Yang, and advised by Michigan SLED Lab Chuang Gan.

Martin Ziqiao Ma

31,958 Aufrufe • vor 11 Tagen

1/ What if you could earn rewards just by learning about crypto? GREED Academy is back with Semester 2, with over 120k in prizes for those who lock, learn, and test their knowledge onchain. Eight weeks. Five sponsors. Big incentives. Time to lock in 📕
0:13

1/ What if you could earn rewards just by learning about crypto? GREED Academy is back with Semester 2, with over 120k in prizes for those who lock, learn, and test their knowledge onchain. Eight weeks. Five sponsors. Big incentives. Time to lock in 📕

GREED Academy 📕

39,360 Aufrufe • vor 1 Jahr

Corey DeAngelis recommending the Federal Gov require the Classic Learning Test as a sane alternative to the SAT and ACT duopoly
0:14

Corey DeAngelis recommending the Federal Gov require the Classic Learning Test as a sane alternative to the SAT and ACT duopoly

Jeremy Wayne Tate

27,669 Aufrufe • vor 9 Monaten