Automated report

deep-diver · Mar 8, 2024 · 523e4ef · 523e4ef
1 parent 89fb85a
commit 523e4ef
Show file tree

Hide file tree

Showing 10 changed files with 102 additions and 0 deletions.
diff --git a/...t/2024-03-08 Chatbot Arena: An Open Platform for Evaluating LLMs by Human Preference.yaml b/...t/2024-03-08 Chatbot Arena: An Open Platform for Evaluating LLMs by Human Preference.yaml
@@ -0,0 +1,9 @@
+date: "2024-03-08"
+author: Wei-Lin Chiang
+title: 'Chatbot Arena: An Open Platform for Evaluating LLMs by Human Preference'
+thumbnail: https://cdn-uploads.huggingface.co/production/uploads/60f1abe7544c2adfd699860c/g-NioYzA0w1stNisWTTJm.png
+link: https://huggingface.co/papers/2403.04132
+summary: This paper presents Chatbot Arena, an open platform that uses human preference to evaluate large language models (LLMs) through a pairwise comparison approach and crowdsourcing. The platform has been operational for several months, collected over 240K votes, and utilizes efficient and accurate statistical methods for evaluation. Chatbot Arena has become a widely cited and referenced LLM leaderboard....
+opinion: placeholder
+tags:
+    - Natural Language Processing
diff --git a/current/2024-03-08 Common 7B Language Models Already Possess Strong Math Capabilities.yaml b/current/2024-03-08 Common 7B Language Models Already Possess Strong Math Capabilities.yaml
@@ -0,0 +1,9 @@
+date: "2024-03-08"
+author: Chen Li
+title: Common 7B Language Models Already Possess Strong Math Capabilities
+thumbnail: https://cdn-uploads.huggingface.co/production/uploads/60f1abe7544c2adfd699860c/eSnCEMNFcNuj00GmG8yur.png
+link: https://huggingface.co/papers/2403.04706
+summary: A language model with common pre-training already has strong mathematical abilities, but its reliability in generating correct answers needs to be improved. Scaling up the synthetic data can significantly enhance its reliability, as it proves to be nearly as effective as real data and shows no clear saturation when scaled up to approximately one million samples....
+opinion: placeholder
+tags:
+    - Natural Language Processing
diff --git a/current/2024-03-08 How Far Are We from Intelligent Visual Deductive Reasoning?.yaml b/current/2024-03-08 How Far Are We from Intelligent Visual Deductive Reasoning?.yaml
@@ -0,0 +1,10 @@
+date: "2024-03-08"
+author: Yizhe Zhang
+title: How Far Are We from Intelligent Visual Deductive Reasoning?
+thumbnail: https://cdn-uploads.huggingface.co/production/uploads/60f1abe7544c2adfd699860c/BXVTNtc2CFDFuU_wvJChb.png
+link: https://huggingface.co/papers/2403.04732
+summary: This paper examines the abilities of Vision-Language Models, like GPT-4, to perform sophisticated visual deductive reasoning using Raven's Progressive Matrices tests. The results show that while these models excel in text-based reasoning, they struggle with visual reasoning and standard strategies used for text-based tasks don't work as well....
+opinion: placeholder
+tags:
+    - Computer Vision
+    - Natural Language Processing
diff --git a/.../2024-03-08 LLMs in the Imaginarium: Tool Learning through Simulated Trial and Error.yaml b/.../2024-03-08 LLMs in the Imaginarium: Tool Learning through Simulated Trial and Error.yaml
@@ -0,0 +1,11 @@
+date: "2024-03-08"
+author: Boshi Wang
+title: 'LLMs in the Imaginarium: Tool Learning through Simulated Trial and Error'
+thumbnail: https://cdn-uploads.huggingface.co/production/uploads/60f1abe7544c2adfd699860c/EtQghz4H4sfuzrxYPinsL.png
+link: https://huggingface.co/papers/2403.04746
+summary: This paper proposes a new method called simulated trial and error (STE) to improve tool use accuracy for large language models (LLMs). STE uses the LLM's imagination to simulate scenarios, memory to improve exploration, and trial and error to learn from feedback. Experiments show that STE improves tool learning for LLMs and outperforms GPT-4....
+opinion: placeholder
+tags:
+    - Supervised Learning
+    - Natural Language Processing
+    - Deep Learning
diff --git a/current/2024-03-08 Pix2Gif: Motion-Guided Diffusion for GIF Generation.yaml b/current/2024-03-08 Pix2Gif: Motion-Guided Diffusion for GIF Generation.yaml
@@ -0,0 +1,12 @@
+date: "2024-03-08"
+author: Hitesh Kandala
+title: 'Pix2Gif: Motion-Guided Diffusion for GIF Generation'
+thumbnail: https://cdn-uploads.huggingface.co/production/uploads/60f1abe7544c2adfd699860c/g3OCXU3aP1Up4xYD-VhQS.png
+link: https://huggingface.co/papers/2403.04634
+summary: The paper presents Pix2Gif, a motion-guided diffusion model for converting images into GIF videos. It uses text and motion prompts to ensure content consistency and coherence, and can be trained on a single node of 16xV100 GPUs. The data and code are publicly available for others to use....
+opinion: placeholder
+tags:
+    - Deep Learning
+    - Computer Vision
+    - Natural Language Processing
+    - Emerging Applications of Machine Learning
diff --git a/...-Σ: Weak-to-Strong Training of Diffusion Transformer for 4K Text-to-Image Generation.yaml b/...-Σ: Weak-to-Strong Training of Diffusion Transformer for 4K Text-to-Image Generation.yaml
@@ -0,0 +1,11 @@
+date: "2024-03-08"
+author: Junsong Chen
+title: 'PixArt-Σ: Weak-to-Strong Training of Diffusion Transformer for 4K Text-to-Image Generation'
+thumbnail: https://cdn-uploads.huggingface.co/production/uploads/60f1abe7544c2adfd699860c/ziRGvTG3Wa3-IxiEklvnw.png
+link: https://huggingface.co/papers/2403.04692
+summary: In this paper, we introduce PixArt-\Sigma, a Diffusion Transformer model~(DiT) capable of directly generating images at 4K resolution. PixArt-\Sigma represents a significant advancement over its predecessor, PixArt-\alpha, offering images of markedly higher fidelity and improved alignment with text prompts. A key feature of PixArt-\Sigma is its training efficiency. Leveraging the foundational pre-training of PixArt-\alpha, it evolves from the `weaker' baseline to a `stronger' model via incorpora...
+opinion: placeholder
+tags:
+    - Deep Learning
+    - Computer Vision
+    - Natural Language Processing
diff --git a/...ent/2024-03-08 Radiative Gaussian Splatting for Efficient X-ray Novel View Synthesis.yaml b/...ent/2024-03-08 Radiative Gaussian Splatting for Efficient X-ray Novel View Synthesis.yaml
@@ -0,0 +1,10 @@
+date: "2024-03-08"
+author: Yuanhao Cai
+title: Radiative Gaussian Splatting for Efficient X-ray Novel View Synthesis
+thumbnail: https://cdn-uploads.huggingface.co/production/uploads/60f1abe7544c2adfd699860c/eEKROluYolawjgMtgxIYr.png
+link: https://huggingface.co/papers/2403.04116
+summary: The paper introduces X-Gaussian, a framework for efficiently generating X-ray images from different viewpoints. It uses a 3D Gaussian splatting-based model and a differentiable radiative rasterization technique to achieve faster training and inference times, outperforming state-of-the-art methods by 6.5 dB. The method has practical applications in sparse-view CT reconstruction and the code and models will be available at <https://github.com/caiyuanhao1998/X-Gaussian>....
+opinion: placeholder
+tags:
+    - Computer Vision
+    - Optimization and Learning Algorithms
diff --git a/current/2024-03-08 StableDrag: Stable Dragging for Point-based Image Editing.yaml b/current/2024-03-08 StableDrag: Stable Dragging for Point-based Image Editing.yaml
@@ -0,0 +1,9 @@
+date: "2024-03-08"
+author: Yutao Cui
+title: 'StableDrag: Stable Dragging for Point-based Image Editing'
+thumbnail: https://cdn-uploads.huggingface.co/production/uploads/60f1abe7544c2adfd699860c/fvJMbDlBIGJycmrI42VOw.png
+link: https://huggingface.co/papers/2403.04437
+summary: The paper presents StableDrag, a framework for stable and precise drag-based image editing. It improves upon existing techniques by introducing a discriminative point tracking method and a confidence-based latent enhancement strategy for motion supervision. This results in more stable and accurate image manipulation, as demonstrated through experiments on DragBench....
+opinion: placeholder
+tags:
+    - Computer Vision
diff --git a/current/2024-03-08 Teaching Large Language Models to Reason with Reinforcement Learning.yaml b/current/2024-03-08 Teaching Large Language Models to Reason with Reinforcement Learning.yaml
@@ -0,0 +1,12 @@
+date: "2024-03-08"
+author: Alex Havrilla
+title: Teaching Large Language Models to Reason with Reinforcement Learning
+thumbnail: https://cdn-uploads.huggingface.co/production/uploads/60f1abe7544c2adfd699860c/wFKm6Ml2h73abCp4v5hsz.png
+link: https://huggingface.co/papers/2403.04642
+summary: The paper investigates the performance of several algorithms for improving large language model reasoning capabilities using reinforcement learning from human feedback. They find that all algorithms perform comparably, with Expert Iteration performing best in most cases and requiring a similar amount of samples as PPO. They also discuss the trade-off between different metrics during fine-tuning and the implications of their findings for the future of LLM fine-tuning with reinforcement learning....
+opinion: placeholder
+tags:
+    - Supervised Learning
+    - Reinforcement Learning
+    - Deep Learning
+    - Natural Language Processing
diff --git a/current/2024-03-08 Yi: Open Foundation Models by 01.AI.yaml b/current/2024-03-08 Yi: Open Foundation Models by 01.AI.yaml
@@ -0,0 +1,9 @@
+date: "2024-03-08"
+author: 01. AI
+title: 'Yi: Open Foundation Models by 01.AI'
+thumbnail: https://cdn-uploads.huggingface.co/production/uploads/60f1abe7544c2adfd699860c/yjdlaxT0u_n0CKGk7fIZ-.png
+link: https://huggingface.co/papers/2403.04652
+summary: This paper introduces the Yi model family, a series of language and multimodal models that demonstrate strong multi-dimensional capabilities. The models are based on large pretrained language models and are extended to chat models, long context models, depth-upscaled models, and vision-language models. The models achieve strong performance on various benchmarks and are attributed to the high quality of the data used for pretraining and finetuning....
+opinion: placeholder
+tags:
+    - Natural Language Processing