Arxiv Papers

This paper investigates the impact of social media influencers on the visibility and citation counts of machine learning research. The study finds that papers endorsed by influencers have significantly higher citation counts, emphasizing the growing influence of social media in scholarly communication. https://arxiv.org/abs//2401.13782 YouTube: https://www.youtube.com/@ArxivPapers TikTok: https://...

The paper proposes a novel pretraining framework called Cross-Attention Masked Autoencoders (CrossMAE) that leverages cross-attention between masked and visible tokens, resulting in improved representation learning with reduced decoding compute. CrossMAE outperforms MAE on ImageNet classification and COCO instance segmentation. https://arxiv.org/abs//2401.14391 YouTube: https://www.youtube.com/@Ar...

The paper proposes a novel pretraining framework called Cross-Attention Masked Autoencoders (CrossMAE) that leverages cross-attention between masked and visible tokens, resulting in improved representation learning with reduced decoding compute. CrossMAE outperforms MAE on ImageNet classification and COCO instance segmentation. https://arxiv.org/abs//2401.14391 YouTube: https://www.youtube.com/@Ar...

This study explores the representation learning abilities of Denoising Diffusion Models (DDM) and finds that only a few modern components are critical for learning good representations, leading to a simplified approach resembling a classical Denoising Autoencoder (DAE). https://arxiv.org/abs//2401.14404 YouTube: https://www.youtube.com/@ArxivPapers TikTok: https://www.tiktok.com/@arxiv_papers Appl...

This study explores the representation learning abilities of Denoising Diffusion Models (DDM) and finds that only a few modern components are critical for learning good representations, leading to a simplified approach resembling a classical Denoising Autoencoder (DAE). https://arxiv.org/abs//2401.14404 YouTube: https://www.youtube.com/@ArxivPapers TikTok: https://www.tiktok.com/@arxiv_papers Appl...

MambaByte, a token-free language model trained on byte sequences, demonstrates computational efficiency and competitive performance compared to subword Transformers, making it a viable option for token-free language modeling. https://arxiv.org/abs//2401.13660 YouTube: https://www.youtube.com/@ArxivPapers TikTok: https://www.tiktok.com/@arxiv_papers Apple Podcasts: https://podcasts.apple.com/us/pod...

MambaByte, a token-free language model trained on byte sequences, demonstrates computational efficiency and competitive performance compared to subword Transformers, making it a viable option for token-free language modeling. https://arxiv.org/abs//2401.13660 YouTube: https://www.youtube.com/@ArxivPapers TikTok: https://www.tiktok.com/@arxiv_papers Apple Podcasts: https://podcasts.apple.com/us/pod...

The paper introduces DITTO, a method for improving the role-playing capabilities of large language models (LLMs) by leveraging their extensive knowledge of characters and dialogues. DITTO outperforms open-source baselines and achieves performance comparable to advanced proprietary chatbots. https://arxiv.org/abs//2401.12474 YouTube: https://www.youtube.com/@ArxivPapers TikTok: https://www.tiktok.c...

The paper introduces DITTO, a method for improving the role-playing capabilities of large language models (LLMs) by leveraging their extensive knowledge of characters and dialogues. DITTO outperforms open-source baselines and achieves performance comparable to advanced proprietary chatbots. https://arxiv.org/abs//2401.12474 YouTube: https://www.youtube.com/@ArxivPapers TikTok: https://www.tiktok.c...

Meta-prompting is a technique that enhances the functionality of language models by transforming them into multi-faceted conductors. It guides the models to break down complex tasks into smaller subtasks, which are handled by expert instances of the same model. This approach improves performance across various tasks and simplifies user interaction. The integration of external tools, such as a Pyth...

Meta-prompting is a technique that enhances the functionality of language models by transforming them into multi-faceted conductors. It guides the models to break down complex tasks into smaller subtasks, which are handled by expert instances of the same model. This approach improves performance across various tasks and simplifies user interaction. The integration of external tools, such as a Pyth...

The paper discusses the use of machine learning algorithms to predict the risk of cardiovascular disease in patients with diabetes. https://arxiv.org/abs//2401.12187 YouTube: https://www.youtube.com/@ArxivPapers TikTok: https://www.tiktok.com/@arxiv_papers Apple Podcasts: https://podcasts.apple.com/us/podcast/arxiv-papers/id1692476016 Spotify: https://podcasters.spotify.com/pod/show/arxiv-papers

The paper discusses the use of machine learning algorithms to predict the risk of cardiovascular disease in patients with diabetes. https://arxiv.org/abs//2401.12187 YouTube: https://www.youtube.com/@ArxivPapers TikTok: https://www.tiktok.com/@arxiv_papers Apple Podcasts: https://podcasts.apple.com/us/podcast/arxiv-papers/id1692476016 Spotify: https://podcasters.spotify.com/pod/show/arxiv-papers

The paper presents a novel method called Binoculars for detecting machine-generated text from large language models (LLMs) without the need for training data. Binoculars achieves high accuracy in detecting machine-generated text across various sources and situations. https://arxiv.org/abs//2401.12070 YouTube: https://www.youtube.com/@ArxivPapers TikTok: https://www.tiktok.com/@arxiv_papers Apple P...

The paper presents a novel method called Binoculars for detecting machine-generated text from large language models (LLMs) without the need for training data. Binoculars achieves high accuracy in detecting machine-generated text across various sources and situations. https://arxiv.org/abs//2401.12070 YouTube: https://www.youtube.com/@ArxivPapers TikTok: https://www.tiktok.com/@arxiv_papers Apple P...

This paper introduces a scheduling strategy for pipeline parallelism in distributed training that achieves zero pipeline bubbles, resulting in improved performance compared to baseline methods. The authors also develop an algorithm to automatically find optimal schedules and introduce a technique to bypass synchronizations during the optimizer step. Experimental evaluations show significant improv...

This paper introduces MEDUSA, a method for improving inference in Large Language Models (LLMs) by adding extra decoding heads to predict multiple tokens in parallel. MEDUSA achieves significant speedup without compromising generation quality. https://arxiv.org/abs//2401.10774 YouTube: https://www.youtube.com/@ArxivPapers TikTok: https://www.tiktok.com/@arxiv_papers Apple Podcasts: https://podcasts...

This paper introduces MEDUSA, a method for improving inference in Large Language Models (LLMs) by adding extra decoding heads to predict multiple tokens in parallel. MEDUSA achieves significant speedup without compromising generation quality. https://arxiv.org/abs//2401.10774 YouTube: https://www.youtube.com/@ArxivPapers TikTok: https://www.tiktok.com/@arxiv_papers Apple Podcasts: https://podcasts...

Pruning up to 20% of parameters in large language models (LLMs) improves their resistance to "Jailbreaking" prompts, reducing the generation of harmful and illegal content without sacrificing performance. Pruning may also enhance other LLM behaviors and improve safety and reliability. https://arxiv.org/abs//2401.10862 YouTube: https://www.youtube.com/@ArxivPapers TikTok: https://www.tikt...

Pruning up to 20% of parameters in large language models (LLMs) improves their resistance to "Jailbreaking" prompts, reducing the generation of harmful and illegal content without sacrificing performance. Pruning may also enhance other LLM behaviors and improve safety and reliability. https://arxiv.org/abs//2401.10862 YouTube: https://www.youtube.com/@ArxivPapers TikTok: https://www.tikt...

The paper introduces a method called Deductive Closure Training (DCT) that uses language models to improve their factuality and coherence. DCT prompts LMs to generate text, reason about its correctness, and fine-tune based on inferred correctness. DCT improves LM fact verification and text generation accuracy. https://arxiv.org/abs//2401.08574 YouTube: https://www.youtube.com/@ArxivPapers TikTok:...

The paper introduces VMamba, a novel architecture that combines the strengths of CNNs and ViTs for visual representation learning. VMamba achieves linear complexity while maintaining global receptive fields and outperforms benchmarks as image resolution increases. https://arxiv.org/abs//2401.10166 YouTube: https://www.youtube.com/@ArxivPapers TikTok: https://www.tiktok.com/@arxiv_papers Apple Podc...

The paper investigates the complexity of training two-layer neural networks with different activation functions. It provides answers to several open questions and establishes the computational hardness of these problems for certain dimensions and activation functions. https://arxiv.org/abs//2303.17045 YouTube: https://www.youtube.com/@ArxivPapers TikTok: https://www.tiktok.com/@arxiv_papers Apple...

The paper investigates the complexity of training two-layer neural networks with different activation functions. It provides answers to several open questions and establishes the computational hardness of these problems for certain dimensions and activation functions. https://arxiv.org/abs//2303.17045 YouTube: https://www.youtube.com/@ArxivPapers TikTok: https://www.tiktok.com/@arxiv_papers Apple...

The paper criticizes the Fréchet Inception Distance (FID) as an evaluation metric for generated images and proposes an alternative metric called CMMD, which is based on CLIP embeddings and offers a more reliable assessment of image quality. https://arxiv.org/abs//2401.09603 YouTube: https://www.youtube.com/@ArxivPapers TikTok: https://www.tiktok.com/@arxiv_papers Apple Podcasts: https://podcasts.a...