Large language models (LLMs) have made significant advancements in the field of natural language processing and are widely applied in tasks such as question answering, dialogue generation, and summarization. However, these models often face the "hallucination" problem, where they generate overly confident but inaccurate information when dealing with uncertain or unknown content, potentially leading to misleading or even harmful outcomes. To address this issue, this paper proposes a self-reflective rollback retrieval-augmented generation (SSR-RAG) method, which balances the accuracy and safety of generated content by integrating external knowledge retrieval and dynamic candidate set optimization. Specifically, we designed a chain retrieval strategy with a rollback mechanism that removes detrimental retrieval results from the candidate set if they negatively impact the final answer generation. Additionally, we introduced a query rewriting module to optimize user input queries and proposed a state evaluation function, PCF, which evaluates the quality of generated content by combining contextual perplexity and output confidence. Experimental results demonstrate that SSR-RAG achieves significant performance improvements on QA datasets such as MMLU, MMCU, and AGIEval, as well as on the safety validation dataset CValues, thereby validating its effectiveness and reliability.

The diffusion model has emerged as a promising approach for image generation and has demonstrated commendable performance in Gesture-Guided Person Image Synthesis (PGPIS). However, existing methods face significant challenges in producing high-quality images with consistent pose alignment. To address these limitations, we propose Hyper-PDT (Hypergraph Prediction Diffusion Transformers), an innovative diffusion model specifically designed for gesture-guided human image generation. Hyper-PDT leverages a unique combination of hypergraph learning and a transformer-based diffusion architecture. The core idea is to utilize hypergraphs to capture high-level feature associations from pose images, enabling the propagation of higher-order information within the semantic space. To enhance the realism of texture details, we introduce a dynamic Classifier-Free Guidance mechanism, which adaptively predicts noise during sampling through dynamic scaling. Extensive quantitative and qualitative evaluations on the DeepFashion benchmark demonstrate that Hyper-PDT significantly outperforms state-of-the-art methods in PGPIS, establishing its superiority in both accuracy and visual quality.

In the era of extensive intersection between art and Artificial Intelligence (AI), such as image generation and fiction co-creation, AI for music remains relatively nascent, particularly in music understanding. This is evident in the limited work on deep music representations, the scarcity of large-scale datasets, and the absence of a universal and community-driven benchmark. To address this issue, we introduce the Music Audio Representation Benchmark for universaL Evaluation, termed MARBLE. It aims to provide a benchmark for various Music Information Retrieval (MIR) tasks by defining a comprehensive taxonomy with four hierarchy levels, including acoustic, performance, score, and high-level description. We then establish a unified protocol based on 18 tasks on 12 public-available datasets, providing a fair and standard assessment of representations of all open-sourced pre-trained models developed on music recordings as baselines. Besides, MARBLE offers an easy-to-use, extendable, and reproducible suite for the community, with a clear statement on copyright issues on datasets. Results suggest recently proposed large-scale pre-trained musical language models perform the best in most tasks, with room for further improvement. The leaderboard and toolkit repository are published to promote future music AI research.