Harnessing Generative AI to Enable Low-Carbon Artificial Intelligence of Things
Generative AI holds immense potential to reduce carbon emissions of Artificial Intelligence of Things (AIoT) through its excellent reasoning and generation capabilities.
Generative AI for Optimizing UAV Communication and Networking Performance
Generative AI can be leveraged to enhance the communication, networking, and security performance of UAV systems by addressing challenges such as dynamic environments, resource constraints, and complex optimization problems.
Reducing Hallucination in Structured Output Generation via Retrieval-Augmented Modeling
Using Retrieval-Augmented Generation (RAG) to reduce hallucination and improve the quality of structured outputs, such as workflows, generated from natural language requirements.
ControlNet++: Enhancing Controllable Text-to-Image Generation with Efficient Pixel-Level Consistency Feedback
ControlNet++ employs pre-trained discriminative reward models to explicitly optimize pixel-level cycle consistency between generated images and input conditional controls, significantly improving controllability without compromising image quality.
Concealing Copyrighted Content in Latent Diffusion Models through Disguised Samples
It is possible to conceal copyrighted images within the training dataset for latent diffusion models by generating disguised samples that are visually distinct from the copyrighted images but share similar latent information.
Responsible Generation of Textual and Visual Content: Addressing Hallucination, Toxicity, and Adversarial Vulnerabilities
Responsible generation of content by generative AI models is crucial for their real-world applicability. This paper investigates the practical responsible requirements of both textual and visual generative models, outlining key considerations such as generating truthful content, avoiding toxic content, refusing harmful instructions, protecting training data privacy, and ensuring generated content identifiability.
Pixel-wise Reinforcement Learning for Aligning Diffusion Models with Human Preferences
Pixel-wise Policy Optimization (PXPO) algorithm that enables diffusion models to receive and optimize for pixel-level feedback from human preferences, improving sample efficiency compared to previous reinforcement learning approaches.
Watermark-based Detection and Attribution of AI-Generated Content: A Systematic Study
Watermark-based detection and attribution is a promising technique to mitigate ethical concerns around generative AI, such as generating harmful content or false copyright claims. This work provides the first systematic study on watermark-based, user-aware detection and attribution of AI-generated content, including theoretical analysis, algorithm development, and extensive empirical evaluation.
Iso-Diffusion: Improving Diffusion Probabilistic Models Using Isotropy of Additive Gaussian Noise
DDPMの信頼性を向上させるために、加法ガウスノイズの等方性を活用する新しい手法を提案します。
SyncTweedies: A General Generative Framework Based on Synchronized Diffusions
SyncTweedies introduces a synchronized diffusion process, SyncTweedies, for generating diverse visual content with superior quality and broad applicability.
© 2024 by Linnk AI