iiinova — where your digital self keeps evolving

SOCIAL MEDIA DESCRIPTION TAG TAG

I think a few things will happen soon: 🚀 Scale beyond 8B 🎯 Multi-modal capabilities ⚡️Faster inference 🔄 Reinforcement learning integration Exciting to see alternatives to autoregressive models succeeding at scale! Paper: ml-gsai.github.io/LLaDA-demo/ (8/8)

ml-gsai.github.io

SOCIAL MEDIA TITLE TAG

000

Results vs LLaMA3 8B: - Matches/exceeds on most tasks - Better at math & Chinese tasks - Strong in-context learning - Improved dialogue capabilities (7/8) 🧵

100

A major result: LLaDA breaks the "reversal curse" that plagues autoregressive models. 🔄 On tasks requiring bidirectional reasoning, it outperforms GPT-4 and maintains consistent performance in both forward/reverse directions. (6/8) 🧵

100

For generation, they introduce clever remasking strategies: - Low-confidence remasking: Remask tokens the model is least sure about - Semi-autoregressive: Generate in blocks left-to-right while maintaining bidirectional context (5/8) 🧵

100

Training uses random masking ratio t ∈ [0,1] for each sequence. The model learns to predict original tokens given partially masked sequences. No causal masking used. Also enables instruction-conditioned generation with the same technique. No modifications. (4/8) 🧵

100

💡Core insight: Generative modeling principles, not autoregression, give LLMs their power. LLaDA's forward process gradually masks tokens while reverse process predicts them simultaneously. This enables bidirectional modeling. (3/8) 🧵

110

Key highlights: - Successful scaling of masked diffusion to LLM scale (8B params) - Masking with variable ratios for forward/reverse process - Smart remasking strategies for generation, incl. semi-autoregressive - SOTA on reversal tasks, matching Llama 3 on others (2/8) 🧵

100

"LLaDA: Large Language Diffusion Models" Nie et al. Just read this fascinating paper. Scaled up Masked Diffusion Language Models to 8B params, and show that it can match #LLMs (including Llama 3) while solving some key limitations! Let's dive in... 🧵 (1/8) #genai

111

Light-A-VideoClick to Play and Loop VideoClick to Play and Loop VideoClick to Play and Loop VideoClick to Play and Loop VideoClick to Play and Loop VideoClick to Play and Loop Video

Project page: bujiazi.github.io/light-a-vide... Code: github.com/bcmi/Light-A... Could be a game-changer for quick video mood/lighting adjustments without complicated VFX pipelines! 🎬

bujiazi.github.io

000

The results are pretty good ✨ They can transform regular videos into moody noir scenes, add sunlight streaming through windows, or create cyberpunk neon vibes -- works on everything from portrait videos to car commercials! 🚗

100

Technical highlights 🔍: - Consistent Light Attention (CLA) module for stable lighting across frames - Progressive Light Fusion for smooth temporal transitions - Works with ANY video diffusion model (AnimateDiff, CogVideoX) - Zero-shot - no fine-tuning needed!

100

New work introduces a training-free method to relight entire videos, while maintaining temporal consistency! 📽️🌅 "Light-A-Video: Training-free Video Relighting via Progressive Light Fusion" Zhou et al. (1/n) 🧵 #genai #ai #research #video

1102

RigAnything: Template-Free Autoregressive Rigging for Diverse 3D Assets

Project page: liuisabella.com/RigAnything/ Code: not available yet Really excited to try this out once the code is available!

liuisabella.com

000

Authors claim that the model generalizes well across diverse shapes - from humanoids to marine creatures! And works with real-world images & arbitrary poses. 🤩

100

Technical highlights: - BFS-ordered skeleton sequence representation - Autoregressive joint prediction with diffusion sampling - Hybrid attention masking: full self-attention for shape tokens, causal attention for skeleton - e2e trainable pipeline without clustering/MST ops

120

Yash Bhalgat @ysbhalgat.bsky.social · 14/02/2025

Need to rig 3D models? 🦖 New work from UCSD and Adobe: "RigAnything: Template-Free Autoregressive Rigging for Diverse 3D Assets" Liu et al. tl;dr: reduces rigging time from 2 mins to 2 secs, works on any shape category & doesn't need predefined templates! 🚀

150

Interesting how they handle the domain gap between 2D latent space and 3D representations through their three-stage pipeline. The correspondence-aware encoding significantly reduces high-frequency noise while preserving geometry. Project: latent-radiance-field.github.io/LRF/

latent-radiance-field.github.io

Latent Radiance Fields with 3D-aware 2D Representations

010

Yash Bhalgat @ysbhalgat.bsky.social · 14/02/2025

Technical approach: - Correspondence-aware autoencoding to enhance 3D consistency in VAE latent space - Builds 3D representations from 3D-aware 2D features - VAE-Radiance Field alignment to bridge domain gap between latent and image space #nerf #ai #research

130

Yash Bhalgat @ysbhalgat.bsky.social · 14/02/2025

"Latent Radiance Fields with 3D-aware 2D Representations" Zhou et al., #ICLR2025 tl;dr: Novel framework that integrates 3D awareness into VAE latent space using correspondence-aware encoding, enabling high-quality rendered images with ~50% memory savings. (1/n) 🧵

120

Yash Bhalgat @ysbhalgat.bsky.social · 13/02/2025

Project: research.nvidia.com/labs/dir/edg... Training and inference code available here: github.com/NVlabs/EdgeR...

000

Yash Bhalgat @ysbhalgat.bsky.social · 13/02/2025

The architecture uses a lightweight encoder and auto-regressive decoder to compress variable-length meshes into fixed-length codes, enabling point cloud and single-image conditioning. Their ArAE model controls face count for varying detail while preserving mesh topology.

100

Yash Bhalgat @ysbhalgat.bsky.social · 13/02/2025

"EdgeRunner" (#ICLR2025) from #Nvidia & PKU introduces an auto-regressive auto-encoder for mesh generation, supporting up to 4000 faces at 512³ resolution. 🤩 Their mesh tokenization algorithm (adapted from EdgeBreaker) achieves ~50% compression (4-5 tokens per face vs 9), making training efficient.

100

Yash Bhalgat @ysbhalgat.bsky.social · 12/02/2025

Technical highlight: They combine 3D latent diffusion with multi-view conditioning for the base shape, then use 2D normal maps for refinement. The results look way cleaner than previous methods.

000

Yash Bhalgat @ysbhalgat.bsky.social · 12/02/2025

Their two-stage approach: First generate coarse geometry (5s), then add fine details (20s) using normal maps based refinement. Smart way to balance speed and quality.

100

Yash Bhalgat @ysbhalgat.bsky.social · 12/02/2025

Just came across this fascinating paper "CraftsMan3D" - a practical approach to text/image-to-3D generation that mimics how artists actually work! Code available (pretrained models too) 🤩: github.com/wyysf-98/Cra... (1/n) 🧵

110

Yash Bhalgat @ysbhalgat.bsky.social · 10/02/2025

Got me excited for a second here 🫠

000

Yash Bhalgat @ysbhalgat.bsky.social · 29/01/2025

So, what happened this week in #AI?

010

(3/n) ⚡️ Speed matters: GSLoc adds just ~180ms overhead while providing substantial accuracy gains. We also provide GSLoc_rel variant for even faster refinement when runtime is critical.

000

(2/n) 📈 Results: GSLoc achieves new SOTA on indoor datasets (7Scenes & 12Scenes) and significantly improves accuracy on Cambridge Landmarks. Our one-shot refinement outperforms methods requiring 50+ optimization steps!

100

(1/n) 🔑 Key idea: We use 3DGS to render high-quality synthetic images & depth maps, enabling efficient one-shot pose refinement of existing APR and SCR methods. No need for iterative optimization or training specialized feature extractors!

100

Yash Bhalgat @ysbhalgat.bsky.social · 21/01/2025

📢 Paper accepted to #ICLR2025 🎉 "GSLoc: Efficient Camera Pose Refinement via 3D Gaussian Splatting" TL;DR: a novel test-time camera pose refinement framework leveraging 3DGS as the scene representation and MASt3R for 2D matching. 🔗: arxiv.org/abs/2408.11085

172

Switzerland is rolling out solar panels... on railway tracks! 🇨🇭🚄 Swiss startup Sun-ways will run a pilot project turning train lines into clean energy highways. #renewable #energy for the win 🤓 www.pv-magazine.com/2024/10/04/s...

pv-magazine.com

Switzerland authorizes removable PV plant on railway track

Swiss startup Sun-ways is planning to build a 18 kW pilot PV system between the racks of a 100-m linear section of a railway line in the Swiss canton of Neuchâtel.

1134

Reposted by Yash Bhalgat

Philip Oldfield @sustainabletall.bsky.social · 13/01/2025

Solar panels are becoming so cheap the Swiss are looking at installing them *between train tracks* !!! www.pv-magazine.com/2024/10/04/s...

38118

Reposted by Yash Bhalgat

Maurice Fallon @mauricefallon.bsky.social · 11/01/2025

🚨🚨🚨 Reminder: closing in 3 weeks time 🚨🚨🚨 Please re-post! Note: Oxford recruits faculty at Associate Professor level - we have no Assistant Professor level.

031

Yash Bhalgat @ysbhalgat.bsky.social · 09/01/2025

Came across this LLM visualisation tool today: bbycroft.net/llm Cool stuff! Let's you visualize each operation or layer in different Transformer architectures, and also explains them on the side. 😍 #llm #visualisation #gpt #ai #transformers

040

Yash Bhalgat @ysbhalgat.bsky.social · 09/01/2025

Didn't know it was that good! Will have to watch it now

000

Reposted by Yash Bhalgat

Michael Niemeyer @miniemeyer.bsky.social · 08/01/2025

For other 3D vision newcomers to blue sky: I highly recommend joning @chrisoffner3d.bsky.social 's list to follow the right people ;): go.bsky.app/Cfm9XFe

2194

Color and aspect-ratio control.

000

(2/2) This method enables inference-time control over SVG attributes like color palettes, aspect ratios, and stroke density, leveraging the learned representation. Really neat approach for editable text-to-vector generation. :)

100

"NeuralSVG: An Implicit Representation for Text-to-Vector Generation" (1/2) Encodes SVGs as implicit neural representations using a small MLP trained with Score Distillation Sampling (SDS). Maps 2D coordinates to shape/color outputs. Dropout-like technique ensures ordered, layered structures.

100