Загрузка видео...

Не удалось загрузить видео

Возникла проблема при загрузке этого видео. Это может быть связано с временными проблемами сети или видео может быть недоступно.

На главную

(1/2) 📢𝐍𝐏𝐆𝐀: 𝐍𝐞𝐮𝐫𝐚𝐥 𝐏𝐚𝐫𝐚𝐦𝐞𝐭𝐫𝐢𝐜 𝐆𝐚𝐮𝐬𝐬𝐢𝐚𝐧 𝐀𝐯𝐚𝐭𝐚𝐫𝐬 📢 #SIGGRAPHAsia We leverage a neural parametric representation to facilitate precise control over 3D Gaussians to obtain high-fidelity avatars.

Matthias Niessner

47,655 subscribers

30,741 просмотров • 1 год назад •via X (Twitter)

Образование Искусство Наука и технологии #SIGGRAPHAsia

Anya Rossi• Live Now

Private livecam show

Комментарии: 0

Нет доступных комментариев

Здесь появятся комментарии из оригинального поста

Похожие видео

(1/2) Check out 𝐆𝐚𝐮𝐬𝐬𝐢𝐚𝐧𝐀𝐯𝐚𝐭𝐚𝐫𝐬: Photorealistic Head Avatars with Rigged 3D Gaussians! We create photorealistic head avatars by animating 3D Gaussians on a parametric face model - edited and rendered in *real-time*!

(1/2) Check out 𝐆𝐚𝐮𝐬𝐬𝐢𝐚𝐧𝐀𝐯𝐚𝐭𝐚𝐫𝐬: Photorealistic Head Avatars with Rigged 3D Gaussians! We create photorealistic head avatars by animating 3D Gaussians on a parametric face model - edited and rendered in real-time!

Matthias Niessner

127,017 просмотров • 2 лет назад

(1/2) We released our Neural Parametric Head Models (NPHM) dataset from our #CVPR2023 paper! It includes over 5600 high-fidelity 3D scans of human heads from 272 subjects - all publicly available! Check it out!

(1/2) We released our Neural Parametric Head Models (NPHM) dataset from our #CVPR2023 paper! It includes over 5600 high-fidelity 3D scans of human heads from 272 subjects - all publicly available! Check it out!

Matthias Niessner

36,064 просмотров • 3 лет назад

(1/2) Intrinsic Image Diffusion for Single-view Material Estimation! We propose a probabilistic diffusion model to handle material & lighting ambiguities. We obtain sharp material estimates and facilitate high-fidelity relighting.

(1/2) Intrinsic Image Diffusion for Single-view Material Estimation! We propose a probabilistic diffusion model to handle material & lighting ambiguities. We obtain sharp material estimates and facilitate high-fidelity relighting.

Matthias Niessner

69,400 просмотров • 2 лет назад

Gaussian Head Avatar: Ultra High-fidelity Head Avatar via Dynamic Gaussians paper page: Creating high-fidelity 3D head avatars has always been a research hotspot, but there remains a great challenge under lightweight sparse view setups. In this paper, we propose Gaussian Head Avatar represented by controllable 3D Gaussians for high-fidelity head avatar modeling. We optimize the neutral 3D Gaussians and a fully learned MLP-based deformation field to capture complex expressions. The two parts benefit each other, thereby our method can model fine-grained dynamic details while ensuring expression accuracy. Furthermore, we devise a well-designed geometry-guided initialization strategy based on implicit SDF and Deep Marching Tetrahedra for the stability and convergence of the training procedure. Experiments show our approach outperforms other state-of-the-art sparse-view methods, achieving ultra high-fidelity rendering quality at 2K resolution even under exaggerated expressions.

AK

65,853 просмотров • 2 лет назад

📢📢📢 Check out XCube, a generative model for very-high resolution 3D shapes and scenes! Project Page: Project Video: Props to Xuanchi Ren for leading this project!!!! [1/N]

📢📢📢 Check out XCube, a generative model for very-high resolution 3D shapes and scenes! Project Page: Project Video: Props to Xuanchi Ren for leading this project!!!! [1/N]

Francis Williams

42,417 просмотров • 2 лет назад

We present “3D magician”: TADA! Text to Animatable Digital Avatars. Given a textual description as input only, our method TADA generates expressive animatable 3D avatars with high-quality geometry and lifelike textures. (1/10)

We present “3D magician”: TADA! Text to Animatable Digital Avatars. Given a textual description as input only, our method TADA generates expressive animatable 3D avatars with high-quality geometry and lifelike textures. (1/10)

Hongwei Yi

52,306 просмотров • 3 лет назад

📢🚗✨ Excited to announce InfiniCube, our scalable generative model for dynamic 3D driving scene generation with high fidelity and controllability! InfiniCube generates very large-scale (300m×400m ~ 100,000m^2), dynamic 3D driving scenes given HD maps, 3D bounding boxes, and text prompts as controls. Website: Props to Yifan Lu for leading this project!!!! [1/N]

📢🚗✨ Excited to announce InfiniCube, our scalable generative model for dynamic 3D driving scene generation with high fidelity and controllability! InfiniCube generates very large-scale (300m×400m ~ 100,000m^2), dynamic 3D driving scenes given HD maps, 3D bounding boxes, and text prompts as controls. Website: Props to Yifan Lu for leading this project!!!! [1/N]

Xuanchi Ren

26,623 просмотров • 1 год назад

When parking, you can now see a high fidelity 3D representation of the world around your vehicle, including proximity & shape of nearby objects, barriers, vehicles & painted road markings By using a dedicated neural network to model obstacles & paint lines, we can accurately estimate distances & represent arbitrary shapes in a smooth & computationally efficient way

When parking, you can now see a high fidelity 3D representation of the world around your vehicle, including proximity & shape of nearby objects, barriers, vehicles & painted road markings By using a dedicated neural network to model obstacles & paint lines, we can accurately estimate distances & represent arbitrary shapes in a smooth & computationally efficient way

Tesla

1,550,383 просмотров • 2 лет назад

Human Hair Reconstruction with Strand-Aligned 3D Gaussians Contributions (cited): – We propose a new 3D line lifting scheme that uses a modified 3DGS reconstruction technique to lift 2D orientation maps into a 3D field while also providing refinement of the camera parameters; – We introduce a dual representation of hair strand polylines and 3D Gaussians to achieve differentiable rasterization of hair strands and leverage photometric constraints for strand-based hair reconstruction; – Based on these components, we propose a coarse-to-fine optimization method for prior-guided hair reconstruction that leverages both latent and explicit representations of the hairstyle.

Human Hair Reconstruction with Strand-Aligned 3D Gaussians Contributions (cited): – We propose a new 3D line lifting scheme that uses a modified 3DGS reconstruction technique to lift 2D orientation maps into a 3D field while also providing refinement of the camera parameters; – We introduce a dual representation of hair strand polylines and 3D Gaussians to achieve differentiable rasterization of hair strands and leverage photometric constraints for strand-based hair reconstruction; – Based on these components, we propose a coarse-to-fine optimization method for prior-guided hair reconstruction that leverages both latent and explicit representations of the hairstyle.

MrNeRF

106,525 просмотров • 1 год назад

📢Pix2NPHM: Learning to Regress NPHM Reconstructions From a Single Image📢 We directly regress neural parametric head models (NPHMs) from a single image — fast, stable, and significantly more expressive than classical 3DMMs such as FLAME. Face tracking & 3D reconstruction are often limited by the representational capacity of PCA-based face models. By lifting NPHMs to a first-class reconstruction primitive, we enable more accurate geometry, richer expressions, and finer animation control. Pix2NPHM obtains fast and reliable NPHM reconstructions on real-world data. Inference-time optimization against surface normals and canonical point maps can further increase fidelity. Key to successful and generalized training of our ViT-based network are: (1) large-scale registration of existing 3D head datasets, and (2) self-supervised training on vast in-the-wild 2D video datasets using pseudo ground-truth surface normals. Finally, we show that geometry-aware pretraining on pixel-aligned reconstruction tasks significantly outperforms generic visual pretraining (e.g., DINO-style features) in terms of generalization. 🌍 🎥 Great work by Simon Giebenhain, Tobias Kirschstein, Liam Schoneveld, Davide Davoli, Zhe Chen

📢Pix2NPHM: Learning to Regress NPHM Reconstructions From a Single Image📢 We directly regress neural parametric head models (NPHMs) from a single image — fast, stable, and significantly more expressive than classical 3DMMs such as FLAME. Face tracking & 3D reconstruction are often limited by the representational capacity of PCA-based face models. By lifting NPHMs to a first-class reconstruction primitive, we enable more accurate geometry, richer expressions, and finer animation control. Pix2NPHM obtains fast and reliable NPHM reconstructions on real-world data. Inference-time optimization against surface normals and canonical point maps can further increase fidelity. Key to successful and generalized training of our ViT-based network are: (1) large-scale registration of existing 3D head datasets, and (2) self-supervised training on vast in-the-wild 2D video datasets using pseudo ground-truth surface normals. Finally, we show that geometry-aware pretraining on pixel-aligned reconstruction tasks significantly outperforms generic visual pretraining (e.g., DINO-style features) in terms of generalization. 🌍 🎥 Great work by Simon Giebenhain, Tobias Kirschstein, Liam Schoneveld, Davide Davoli, Zhe Chen

Matthias Niessner

37,850 просмотров • 7 месяцев назад

Wonderland: Navigating 3D Scenes from a Single Image Contributions: • First, we introduce a representation for controllable 3D generation by leveraging the generative priors from camera-guided video diffusion models. Unlike image models, video diffusion models are trained on extensive video datasets. This enables them to capture comprehensive spatial relationships within scenes across multiple views and embed a form of "3D awareness" in their latent space, which allows us to maintain 3D consistency in novel view synthesis. • Second, to achieve controllable novel view generation, we empower video models with precise control over specified camera motions. We introduce a novel dual-branch conditioning mechanism that effectively incorporates desired diverse camera trajectories into the video diffusion model. This enables expansion of a single image into a multi-view consistent capture of a 3D scene with precise pose control. • Third, to achieve efficient 3D reconstruction, we directly transform video latents into 3DGS. We propose a novel latent-based large reconstruction model (LaLRM) that lifts video latents to 3D in a feed-forward manner. With this design, during inference, our model directly predicts 3DGS from a single input image, effectively aligning the generation and reconstruction tasks—and bridging image space and 3D space—through the video latent space. Compared with reconstructing scenes from images, the video latent space offers a 256× spatial-temporal reduction while retaining essential and consistent 3D structural details. Such a high degree of compression is crucial, as it allows the LaLRM to handle a wider range of 3D scenes within the reconstruction framework, with the same memory constraints.

Wonderland: Navigating 3D Scenes from a Single Image Contributions: • First, we introduce a representation for controllable 3D generation by leveraging the generative priors from camera-guided video diffusion models. Unlike image models, video diffusion models are trained on extensive video datasets. This enables them to capture comprehensive spatial relationships within scenes across multiple views and embed a form of "3D awareness" in their latent space, which allows us to maintain 3D consistency in novel view synthesis. • Second, to achieve controllable novel view generation, we empower video models with precise control over specified camera motions. We introduce a novel dual-branch conditioning mechanism that effectively incorporates desired diverse camera trajectories into the video diffusion model. This enables expansion of a single image into a multi-view consistent capture of a 3D scene with precise pose control. • Third, to achieve efficient 3D reconstruction, we directly transform video latents into 3DGS. We propose a novel latent-based large reconstruction model (LaLRM) that lifts video latents to 3D in a feed-forward manner. With this design, during inference, our model directly predicts 3DGS from a single input image, effectively aligning the generation and reconstruction tasks—and bridging image space and 3D space—through the video latent space. Compared with reconstructing scenes from images, the video latent space offers a 256× spatial-temporal reduction while retaining essential and consistent 3D structural details. Such a high degree of compression is crucial, as it allows the LaLRM to handle a wider range of 3D scenes within the reconstruction framework, with the same memory constraints.

MrNeRF

52,849 просмотров • 1 год назад

📢GeomHair: Reconstruction of Hair Strands from Colorless 3D Scans📢 We present a novel method to reconstruct hair strands from colorless 3D scans by extracting orientation cues directly from the mesh surface geometry by finding local characteristic lines and from shaded renderings using a neural 2D line detector. We enhance the reconstruction with a diffusion prior trained on synthetic hair data and adapted to each scan using a tailored text prompt, allowing us to recover both simple and complex hairstyles without relying on color input. To support further research, we also introduce Strands400, the largest publicly available dataset of 3D hair strand reconstructions from real-world scans of 400 different people, featuring complicated hairstyles, such as ponytails and buns. 🌍 📷 Great work by Rachmadio Noval L. Artem Sevastopolsky Egor Zakharov @ness_pris

📢GeomHair: Reconstruction of Hair Strands from Colorless 3D Scans📢 We present a novel method to reconstruct hair strands from colorless 3D scans by extracting orientation cues directly from the mesh surface geometry by finding local characteristic lines and from shaded renderings using a neural 2D line detector. We enhance the reconstruction with a diffusion prior trained on synthetic hair data and adapted to each scan using a tailored text prompt, allowing us to recover both simple and complex hairstyles without relying on color input. To support further research, we also introduce Strands400, the largest publicly available dataset of 3D hair strand reconstructions from real-world scans of 400 different people, featuring complicated hairstyles, such as ponytails and buns. 🌍 📷 Great work by Rachmadio Noval L. Artem Sevastopolsky Egor Zakharov @ness_pris

Matthias Niessner

12,490 просмотров • 1 год назад

3D Gaussian Splatting for Real-Time Radiance Field Rendering paper page: Radiance Field methods have recently revolutionized novel-view synthesis of scenes captured with multiple photos or videos. However, achieving high visual quality still requires neural networks that are costly to train and render, while recent faster methods inevitably trade off speed for quality. For unbounded and complete scenes (rather than isolated objects) and 1080p resolution rendering, no current method can achieve real-time display rates. We introduce three key elements that allow us to achieve state-of-the-art visual quality while maintaining competitive training times and importantly allow high-quality real-time (>= 30 fps) novel-view synthesis at 1080p resolution. First, starting from sparse points produced during camera calibration, we represent the scene with 3D Gaussians that preserve desirable properties of continuous volumetric radiance fields for scene optimization while avoiding unnecessary computation in empty space; Second, we perform interleaved optimization/density control of the 3D Gaussians, notably optimizing anisotropic covariance to achieve an accurate representation of the scene; Third, we develop a fast visibility-aware rendering algorithm that supports anisotropic splatting and both accelerates training and allows realtime rendering. We demonstrate state-of-the-art visual quality and real-time rendering on several established datasets.

3D Gaussian Splatting for Real-Time Radiance Field Rendering paper page: Radiance Field methods have recently revolutionized novel-view synthesis of scenes captured with multiple photos or videos. However, achieving high visual quality still requires neural networks that are costly to train and render, while recent faster methods inevitably trade off speed for quality. For unbounded and complete scenes (rather than isolated objects) and 1080p resolution rendering, no current method can achieve real-time display rates. We introduce three key elements that allow us to achieve state-of-the-art visual quality while maintaining competitive training times and importantly allow high-quality real-time (>= 30 fps) novel-view synthesis at 1080p resolution. First, starting from sparse points produced during camera calibration, we represent the scene with 3D Gaussians that preserve desirable properties of continuous volumetric radiance fields for scene optimization while avoiding unnecessary computation in empty space; Second, we perform interleaved optimization/density control of the 3D Gaussians, notably optimizing anisotropic covariance to achieve an accurate representation of the scene; Third, we develop a fast visibility-aware rendering algorithm that supports anisotropic splatting and both accelerates training and allows realtime rendering. We demonstrate state-of-the-art visual quality and real-time rendering on several established datasets.

AK

633,674 просмотров • 3 лет назад

In the next update we will add a voting feature to our staking platform We are planning to launch the first voting on Sunday and invite XRPL projects to take part in it. 📢Tag project with the strongest community and we will invite them to vote! (1/2)⬇️

In the next update we will add a voting feature to our staking platform We are planning to launch the first voting on Sunday and invite XRPL projects to take part in it. 📢Tag project with the strongest community and we will invite them to vote! (1/2)⬇️

Rocket Launch

19,610 просмотров • 3 лет назад

1/2 📢🎉🎉 $KIMA Futures are LIVE on MEXC! 🎉🎉📢 We're absolutely thrilled to share that $KIMA is now LIVE on MEXC Futures! 🤩Why is this a big deal? Trading futures on $KIMA gives you the chance to amplify your trading potential with up to 50x leverage! ⚡️More buying power: Control positions much larger than your initial investment. ⚡️Maximized opportunities: Ride the $KIMA momentum with enhanced market exposure. 😮 What’s in it for you? ⚡️Access to the dynamic $KIMA market with cutting-edge futures trading tools. ⚡️Possibilities to elevate your portfolio like never before. 😉Ready to dive in? ➡️Start trading $KIMA on MEXC futures now:

1/2 📢🎉🎉 $KIMA Futures are LIVE on MEXC! 🎉🎉📢 We're absolutely thrilled to share that $KIMA is now LIVE on MEXC Futures! 🤩Why is this a big deal? Trading futures on $KIMA gives you the chance to amplify your trading potential with up to 50x leverage! ⚡️More buying power: Control positions much larger than your initial investment. ⚡️Maximized opportunities: Ride the $KIMA momentum with enhanced market exposure. 😮 What’s in it for you? ⚡️Access to the dynamic $KIMA market with cutting-edge futures trading tools. ⚡️Possibilities to elevate your portfolio like never before. 😉Ready to dive in? ➡️Start trading $KIMA on MEXC futures now:

Kima Network

32,794 просмотров • 1 год назад

📢📢𝐍𝐞𝐑𝐒𝐞𝐦𝐛𝐥𝐞 𝐯𝟐 𝐃𝐚𝐭𝐚𝐬𝐞𝐭 𝐑𝐞𝐥𝐞𝐚𝐬𝐞📢📢 Head captures of 7.1MP from 16 cameras at 73fps: * More recordings (425 people) * Better color calibration * Convenient download scripts The new version of our dataset adds 156 participants for a total of 425 different people. In its entirety, the dataset provides now 65 million images from over 15 hours of diverse human facial expression performances. We improved the color consistency of the recorded images with a better color calibration procedure. As a result, 3D reconstructions with images from the NeRSemble dataset should become better and look more realistic. Finally, we made it much easier to download the recordings with our new download repository. It now just takes a single command to download all frontal hair shake videos of all participants or to download all recordings of a single participant. Check it out: Awesome work by Tobias Kirschstein Simon Giebenhain !!!

📢📢𝐍𝐞𝐑𝐒𝐞𝐦𝐛𝐥𝐞 𝐯𝟐 𝐃𝐚𝐭𝐚𝐬𝐞𝐭 𝐑𝐞𝐥𝐞𝐚𝐬𝐞📢📢 Head captures of 7.1MP from 16 cameras at 73fps: * More recordings (425 people) * Better color calibration * Convenient download scripts The new version of our dataset adds 156 participants for a total of 425 different people. In its entirety, the dataset provides now 65 million images from over 15 hours of diverse human facial expression performances. We improved the color consistency of the recorded images with a better color calibration procedure. As a result, 3D reconstructions with images from the NeRSemble dataset should become better and look more realistic. Finally, we made it much easier to download the recordings with our new download repository. It now just takes a single command to download all frontal hair shake videos of all participants or to download all recordings of a single participant. Check it out: Awesome work by Tobias Kirschstein Simon Giebenhain !!!

Matthias Niessner

12,089 просмотров • 1 год назад

Beyond generating high-fidelity visuals, we wanted to test the limits of what Nano Banana Pro can do. We worked with design partners Porto Rocha to build out a hypothetical brand called YOYOYO to see how the model would handle the task. Here’s what we found: 🎨Brand consistency: Across logos, colors, and typography, the model maintained a strict, cohesive brand identity (even for wildly diverse concepts) 🛍️Environmental realism: We asked to see the products in storefront and studio mockups. It nailed accurate lighting, shadows, and physical proportions - even when upscaled for massive retail displays 🪀Spatial accuracy: We tested spatial volumes for physical packaging. The generated proportions were so precise that we were able to 3D-print the functional yo-yo How have you been pushing the limits of Nano Banana Pro? Let us know in the replies below!

Google AI

123,471 просмотров • 3 месяцев назад

✨ Made a new mini feature on Photo AI: [ Grab from 3d model ] So the problem is we're at that stage in time (typical for AI) where image-to-3d models are not good enough but are fun to play with, but we know they'll be good enough in 1-2 years With [ Make 3d model ] you already can turn any Photo AI pic into a 3d model but it still looks hyper clunky and deformed, but it works! One cool idea I had to make that more useful and made now: Let people make a 3d model then change the view of the it with the 3d viewer, then press [ o ] and it grabs a frame of the 3d That image you can then [ Remix ] (img2img), and it becomes a real photo again and that in turn you can then turn into a video again with [ Make video ] So that essentially gives you a fully freeform camera position control to take photos with One thing I need to fix is the background/skybox, I kinda need to take the original photo and remove the person and just get the background for the 3d model viewer, in this case it should be white, but it's a start!

✨ Made a new mini feature on Photo AI: [ Grab from 3d model ] So the problem is we're at that stage in time (typical for AI) where image-to-3d models are not good enough but are fun to play with, but we know they'll be good enough in 1-2 years With [ Make 3d model ] you already can turn any Photo AI pic into a 3d model but it still looks hyper clunky and deformed, but it works! One cool idea I had to make that more useful and made now: Let people make a 3d model then change the view of the it with the 3d viewer, then press [ o ] and it grabs a frame of the 3d That image you can then [ Remix ] (img2img), and it becomes a real photo again and that in turn you can then turn into a video again with [ Make video ] So that essentially gives you a fully freeform camera position control to take photos with One thing I need to fix is the background/skybox, I kinda need to take the original photo and remove the person and just get the background for the 3d model viewer, in this case it should be white, but it's a start!

@levelsio

119,210 просмотров • 1 год назад

GIVEAWAY ALERT!!! 🔴📢📢 this one's gonna be good! for this giveaway we are offering 1 metal gspot complete kit as well as 1 super safety 2 pos pa6cf printed kit and 1 nylon sloppy seconds kit, that's 3 winners! each winner will also receive a custom laser etched ash tray courtesy of BigSmellyWabo! so make sure you hit him up for custom work. RULES FOR ENTRY: simply follow us as well as like and repost, must be 18 or older and reside in the continental US! winners will be picked 8/1/2025 at 12pm cst, winners will be chosen randomly, no extra entries per shares etc. good luck to you guys, and if your interested in what we do check us out at Rusty forks

Sensitive content

GIVEAWAY ALERT!!! 🔴📢📢 this one's gonna be good! for this giveaway we are offering 1 metal gspot complete kit as well as 1 super safety 2 pos pa6cf printed kit and 1 nylon sloppy seconds kit, that's 3 winners! each winner will also receive a custom laser etched ash tray courtesy of BigSmellyWabo! so make sure you hit him up for custom work. RULES FOR ENTRY: simply follow us as well as like and repost, must be 18 or older and reside in the continental US! winners will be picked 8/1/2025 at 12pm cst, winners will be chosen randomly, no extra entries per shares etc. good luck to you guys, and if your interested in what we do check us out at Rusty forks

Pembleton & sons'

25,002 просмотров • 1 год назад

[LSTM] by Hand ✍️ LSTMs have been the most effective architecture to process long sequences of data, until our world was taken over by the Transformers. LSTMs belong to the broader family of recurrent neural network (RNNs) that process data sequentially in a recurrent manner. Transformers, on the other hand, abandon recurrence and use self-attention instead to process data concurrently in parallel. Recently, there is renewed interest in recurrence as people realized self-attention doesn’t scale to extremely long sequences, like hundreds of thousands of tokens. Mamba is a good example to bring back recurrence. All of a sudden, it is cool to study LSTMs. How do LSTMs work? [1] Given ↳ 🟨 Input sequence X1, X2, X3 (d = 3) ↳ 🟩 Hidden state h (d = 2) ↳ 🟦 Memory C (d = 2) ↳ Weight matrices Wf, Wc, Wi, Wo Process t = 1 [2] Initialize ↳ Randomly set the previous hidden state h0 to [1, 1] and memory cells C0 to [0.3, -0.5] [3] Linear Transform ↳ Multiply the four weight matrices with the concatenation of current input (X1) and the previous hidden state (h0). ↳ The results are feature values, each is a linear combination of the current input and hidden state. [4] Non-linear Transform ↳ Apply sigmoid σ to obtain gate values (between 0 and 1). • Forget gate (f1): [-4, -6] → [0, 0] • Input gate (i1): [6, 4] → [1, 1] • Output gate (o1): [4, -5] → [1, 0] ↳ Apply tanh to obtain candidate memory values (between -1 and 1) • Candidate memory (C’1): [1, -6] → [0.8, -1] [5] Update Memory ↳ Forget (C0 .* f1): Element-wise multiply the current memory with forget gate values. ↳ Input (C’1 .* o1): Element-wise multiply the “candidate” memory with input gate values. ↳ Update the memory to C1 by adding the two terms above: C0 .* f1 + C’1 .* o1 = C1 [6] Candiate Output ↳ Apply tanh to the new memory C1 to obtain candidate output o’1. [0.8, -1] → [0.7, -0.8] [7] Update Hidden State ↳ Output (o’1 .* o1 → h1): Element-wise multiply the candidate output with the output gate. ↳ The result is updated hidden state h1 ↳ Also, it is the first output. Process t = 2 [8] Initialize ↳ Copy previous hidden state h1 and memory C1 [9] Linear Transform ↳ Repeat [3] [10] Update Memory (C2) ↳ Repeat [4] and [5] [11] Update Hidden State (h2) ↳ Repeat [6] and [7] Process t = 3 [12] Initialize ↳ Copy previous hidden state h2 and memory C2 [13] Linear Transform ↳ Repeat [3] [14] Update Memory (C3) ↳ Repeat [4] and [5] [15] Update Hidden State (h3) ↳ Repeat [6] and [7]

[LSTM] by Hand ✍️ LSTMs have been the most effective architecture to process long sequences of data, until our world was taken over by the Transformers. LSTMs belong to the broader family of recurrent neural network (RNNs) that process data sequentially in a recurrent manner. Transformers, on the other hand, abandon recurrence and use self-attention instead to process data concurrently in parallel. Recently, there is renewed interest in recurrence as people realized self-attention doesn’t scale to extremely long sequences, like hundreds of thousands of tokens. Mamba is a good example to bring back recurrence. All of a sudden, it is cool to study LSTMs. How do LSTMs work? [1] Given ↳ 🟨 Input sequence X1, X2, X3 (d = 3) ↳ 🟩 Hidden state h (d = 2) ↳ 🟦 Memory C (d = 2) ↳ Weight matrices Wf, Wc, Wi, Wo Process t = 1 [2] Initialize ↳ Randomly set the previous hidden state h0 to [1, 1] and memory cells C0 to [0.3, -0.5] [3] Linear Transform ↳ Multiply the four weight matrices with the concatenation of current input (X1) and the previous hidden state (h0). ↳ The results are feature values, each is a linear combination of the current input and hidden state. [4] Non-linear Transform ↳ Apply sigmoid σ to obtain gate values (between 0 and 1). • Forget gate (f1): [-4, -6] → [0, 0] • Input gate (i1): [6, 4] → [1, 1] • Output gate (o1): [4, -5] → [1, 0] ↳ Apply tanh to obtain candidate memory values (between -1 and 1) • Candidate memory (C’1): [1, -6] → [0.8, -1] [5] Update Memory ↳ Forget (C0 .* f1): Element-wise multiply the current memory with forget gate values. ↳ Input (C’1 .* o1): Element-wise multiply the “candidate” memory with input gate values. ↳ Update the memory to C1 by adding the two terms above: C0 .* f1 + C’1 .* o1 = C1 [6] Candiate Output ↳ Apply tanh to the new memory C1 to obtain candidate output o’1. [0.8, -1] → [0.7, -0.8] [7] Update Hidden State ↳ Output (o’1 .* o1 → h1): Element-wise multiply the candidate output with the output gate. ↳ The result is updated hidden state h1 ↳ Also, it is the first output. Process t = 2 [8] Initialize ↳ Copy previous hidden state h1 and memory C1 [9] Linear Transform ↳ Repeat [3] [10] Update Memory (C2) ↳ Repeat [4] and [5] [11] Update Hidden State (h2) ↳ Repeat [6] and [7] Process t = 3 [12] Initialize ↳ Copy previous hidden state h2 and memory C2 [13] Linear Transform ↳ Repeat [3] [14] Update Memory (C3) ↳ Repeat [4] and [5] [15] Update Hidden State (h3) ↳ Repeat [6] and [7]

Tom Yeh

72,966 просмотров • 2 лет назад