fly51fly @fly51fly

BUPT prof | Sharing latest AI papers & insights | Join me in embracing the AI revolution! #MachineLearning #AI #Innovation github.com/fly51fly Joined February 2009

Tweets

23K
Followers

5K
Following

2K
Likes

77K

fly51fly @fly51fly

6 hours ago

[LG] AdaQAT: Adaptive Bit-Width Quantization-Aware Training C Gernigon, S Filip, O Sentieys… [Inria] (2024) arxiv.org/abs/2404.16876 - AdaQAT is an optimization-based method for mixed-precision uniform quantization of both weights and activations in DNNs. - It uses relaxed…

1 1 3 607 3

Download Image

fly51fly @fly51fly

7 hours ago

[CL] Examining the robustness of LLM evaluation to the distributional assumptions of benchmarks M Ailem, K Marazopoulou, C Siska, J Bono [Microsoft] (2024) arxiv.org/abs/2404.16966 - The standard approach for evaluating LLMs on benchmarks assumes test prompts represent a random…

1 3 11 1K 6

Download Image

fly51fly @fly51fly

7 hours ago

[CL] Player-Driven Emergence in LLM-Driven Game Narrative X Peng, J Quaye, W Xu, C Brockett… [Microsoft Research] (2024) arxiv.org/abs/2404.17027 - Players interact with non-deterministic NPCs generated by GPT-4 in a text-adventure game where they try to solve a mystery and…

1 0 3 518 5

Download Image

fly51fly @fly51fly

21 hours ago

[CL] Preference Fine-Tuning of LLMs Should Leverage Suboptimal, On-Policy Data arxiv.org/abs/2404.14367 - This paper aims to understand the behaviors of various procedures for fine-tuning language models with preference data, including RL, maximum likelihood, and…

2 6 24 3K 21

Download Image

fly51fly @fly51fly

21 hours ago

[LG] Parameter-Efficient Fine-Tuning for Large Models: A Comprehensive Survey arxiv.org/abs/2403.14608 - Large models have achieved remarkable performance but require substantial computational resources for fine-tuning. Parameter Efficient Fine-Tuning (PEFT) provides a…

1 13 37 3K 23

Download Image

fly51fly @fly51fly

21 hours ago

[LG] A Survey on the Memory Mechanism of Large Language Model based Agents arxiv.org/abs/2404.13501 - The memory module is a key component that differentiates agents from original large language models (LLMs), enabling agent-environment interactions. - Memory serves…

1 32 78 9K 89

Download Image

fly51fly @fly51fly

21 hours ago

[CV] How Far Are We to GPT-4V? Closing the Gap to Commercial Multimodal Models with Open-Source Suites arxiv.org/abs/2404.16821 - InternVL 1.5 is an open-source multimodal large language model that aims to bridge the capability gap between open-source and proprietary…

1 9 24 2K 15

Download Image

fly51fly @fly51fly

21 hours ago

[CV] Align Your Steps: Optimizing Sampling Schedules in Diffusion Models A Sabour, S Fidler, K Kreis [NVIDIA] (2024) arxiv.org/abs/2404.14507 - Diffusion models have become state-of-the-art generative models, but are slow to sample from due to their reliance on sequential…

0 8 16 2K 11

Download Image

fly51fly @fly51fly

22 hours ago

[CV] NeRF-XL: Scaling NeRFs with Multiple GPUs R Li, S Fidler, A Kanazawa, F Williams [NVIDIA & UC Berkeley] (2024) arxiv.org/abs/2404.16221 - The paper introduces NeRF-XL, a method for distributing Neural Radiance Fields (NeRFs) across multiple GPUs to enable training and…

1 4 10 1K 6

Download Image

fly51fly @fly51fly

a day ago

[AS] Music Consistency Models Z Fei, M Fan, J Huang [Kunlun Inc] (2024) arxiv.org/abs/2404.13358 - Proposes Music Consistency Models (MusicCM) which leverages consistency models for efficient high-quality music synthesis with minimal sampling steps. - Builds on existing…

1 6 18 2K 8

Download Image

fly51fly @fly51fly

a day ago

[CL] LLM Evaluators Recognize and Favor Their Own Generations A Panickssery, S R. Bowman, S Feng [MATS & New York University] (2024) arxiv.org/abs/2404.13076 - Frontier LLMs like GPT-3.5, GPT-4, and Llama 2 exhibit self-preference when evaluating their own text summaries versus…

2 5 30 3K 16

Download Image

fly51fly @fly51fly

a day ago

[CL] FlowMind: Automatic Workflow Generation with LLMs Z Zeng, W Watson, N Cho, S Rahimi… [J. P. Morgan AI Research] (2024) arxiv.org/abs/2404.13050 - FlowMind is a novel approach that uses LLMs to automatically generate workflows, addressing issues like hallucination and data…

1 6 11 1K 8

Download Image

fly51fly @fly51fly

2 days ago

[CV] Editable Image Elements for Controllable Synthesis arxiv.org/abs/2404.16029 - The paper proposes a new image representation called "image elements" that supports spatial editing of input images and faithful reconstruction. - Image elements are obtained by dividing…

0 1 10 1K 7

Download Image

fly51fly @fly51fly

2 days ago

[CV] MoDE: CLIP Data Experts via Clustering arxiv.org/abs/2404.16030 - The noisy pairing in web-crawled image-caption data hurts the learning of CLIP via contrastive pretraining. Semantically different captions for similar images introduce false negatives which hurts…

1 17 49 5K 28

Download Image

fly51fly @fly51fly

2 days ago

[CV] CatLIP: CLIP-level Visual Recognition Accuracy with 2.7x Faster Pre-training on Web-scale Image-Text Data arxiv.org/abs/2404.15653 - CatLIP refines pre-training on image-text data as an image classification task instead of contrastive learning. This removes the need for…

0 6 19 2K 8

Download Image

fly51fly @fly51fly

2 days ago

[CL] How Good Are Low-bit Quantized LLaMA3 Models? An Empirical Study W Huang, X Ma, H Qin, X Zheng... [The University of Hong Kong & Beihang University & ETH Zurich] (2024) arxiv.org/abs/2404.14047 - LLAMA3 models recently released by Meta have achieved state-of-the-art…

1 10 18 2K 10

Download Image

fly51fly @fly51fly

2 days ago

[CV] RadRotator: 3D Rotation of Radiographs with Diffusion Models P Rouzrokh, B Khosravi, S Faghani, K L. Mulford, M J. Taunton, B J. Erickson, C C. Wyles [Mayo Clinic] (2024) arxiv.org/abs/2404.13000 - The paper introduces a diffusion model-based method to rotate the anatomical…

0 1 3 1K 3

Download Image

fly51fly @fly51fly

2 days ago

[LG] Graph Machine Learning in the Era of Large Language Models (LLMs) W Fan, S Wang, J Huang, Z Chen… [he Hong Kong Polytechnic University] (2024) arxiv.org/abs/2404.14928 - GNNs have become a cornerstone in Graph ML for modeling complex relationships in graph data across…

1 8 23 2K 10

Download Image

fly51fly @fly51fly

2 days ago

[CL] LongEmbed: Extending Embedding Models for Long Context Retrieval D Zhu, L Wang, N Yang, Y Song, W Wu, F Wei, S Li [Microsoft & Peking University] (2024) arxiv.org/abs/2404.12096 - Long context embedding models have been confined to a narrow context window not exceeding 8k…

1 5 22 2K 13

Download Image

fly51fly @fly51fly

2 days ago

[CL] SpaceByte: Towards Deleting Tokenization from Large Language Modeling K Slagle [Rice University] (2024) arxiv.org/abs/2404.14408 - Tokenization improves performance of large LMs but has disadvantages like biases, adversarial vulnerability, worse character modeling, and…