In on-device Retrieval Augmented Generation (RAG) systems, the limitations of device computational capabilities and data privacy restrict the use of powerful models, such as large language models and advanced text embedding models, necessitating reliance on smaller alternatives. Consequently, currently used pipelines heavily rely on LLMs for a comprehensive understanding of text semantics when computing embedding similarity for retrieval, facing significant challenges. These smaller models often struggle to capture the precise semantic nuances within lengthy texts, complicating accurate matching. To tackle these challenges, it is essential to: i) Reduce the complexity of input content for generation, ensuring that semantic information is clear and concise; ii) Shorten the length of input content for smaller language models, facilitating improved comprehension and retrieval accuracy. Additionally, employing effective graph indexing structures can help mitigate performance deficiencies in semantic matching, thereby enhancing the overall retrieval process. In MiniRAG, we propose a Graph-based Knowledge Retrieval mechanism that effectively leverages the semantic-aware heterogeneous graph G constructed during the indexing phase, in conjunction with lightweight text embeddings, to achieve efficient knowledge retrieval. By employing a graph-based search design, we aim to ease the burden on precise semantic matching with large language models. This approach facilitates the acquisition of rich and accurate textual content at a low computational cost, thereby enhancing the ability of language models to generate precise responses.

We propose the Cooperative Aerial Robot Inspection Challenge (CARIC), a simulation-based benchmark for motion planning algorithms in heterogeneous multi-UAV systems. CARIC features UAV teams with complementary sensors, realistic constraints, and evaluation metrics prioritizing inspection quality and efficiency. It offers a ready-to-use perception-control software stack and diverse scenarios to support the development and evaluation of task allocation and motion planning algorithms. Competitions using CARIC were held at IEEE CDC 2023 and the IROS 2024 Workshop on Multi-Robot Perception and Navigation, attracting innovative solutions from research teams worldwide. This paper examines the top three teams from CDC 2023, analyzing their exploration, inspection, and task allocation strategies while drawing insights into their performance across scenarios. The results highlight the task's complexity and suggest promising directions for future research in cooperative multi-UAV systems.

Lyric-to-melody generation is a highly challenging task in the field of AI music generation. Due to the difficulty of learning strict yet weak correlations between lyrics and melodies, previous methods have suffered from weak controllability, low-quality and poorly structured generation. To address these challenges, we propose CSL-L2M, a controllable song-level lyric-to-melody generation method based on an in-attention Transformer decoder with fine-grained lyric and musical controls, which is able to generate full-song melodies matched with the given lyrics and user-specified musical attributes. Specifically, we first introduce REMI-Aligned, a novel music representation that incorporates strict syllable- and sentence-level alignments between lyrics and melodies, facilitating precise alignment modeling. Subsequently, sentence-level semantic lyric embeddings independently extracted from a sentence-wise Transformer encoder are combined with word-level part-of-speech embeddings and syllable-level tone embeddings as fine-grained controls to enhance the controllability of lyrics over melody generation. Then we introduce human-labeled musical tags, sentence-level statistical musical attributes, and learned musical features extracted from a pre-trained VQ-VAE as coarse-grained, fine-grained and high-fidelity controls, respectively, to the generation process, thereby enabling user control over melody generation. Finally, an in-attention Transformer decoder technique is leveraged to exert fine-grained control over the full-song melody generation with the aforementioned lyric and musical conditions. Experimental results demonstrate that our proposed CSL-L2M outperforms the state-of-the-art models, generating melodies with higher quality, better controllability and enhanced structure. Demos and source code are available at https://lichaiustc.github.io/CSL-L2M/.

Musical mode is one of the most critical element that establishes the framework of pitch organization and determines the harmonic relationships. Previous works often use the simplistic and rigid alignment method, and overlook the diversity of modes. However, in contrast to AI models, humans possess cognitive mechanisms for perceiving the various modes and keys. In this paper, we propose a spiking neural network inspired by brain mechanisms and psychological theories to represent musical modes and keys, ultimately generating musical pieces that incorporate tonality features. Specifically, the contributions are detailed as follows: 1) The model is designed with multiple collaborated subsystems inspired by the structures and functions of corresponding brain regions; 2)We incorporate mechanisms for neural circuit evolutionary learning that enable the network to learn and generate mode-related features in music, reflecting the cognitive processes involved in human music perception. 3)The results demonstrate that the proposed model shows a connection framework closely similar to the Krumhansl-Schmuckler model, which is one of the most significant key perception models in the music psychology domain. 4) Experiments show that the model can generate music pieces with characteristics of the given modes and keys. Additionally, the quantitative assessments of generated pieces reveals that the generating music pieces have both tonality characteristics and the melodic adaptability needed to generate diverse and musical content. By combining insights from neuroscience, psychology, and music theory with advanced neural network architectures, our research aims to create a system that not only learns and generates music but also bridges the gap between human cognition and artificial intelligence.

The rapid advancement of Large Vision-Language Models (LVLMs) has enhanced capabilities offering potential applications from content creation to productivity enhancement. Despite their innovative potential, LVLMs exhibit vulnerabilities, especially in generating potentially toxic or unsafe responses. Malicious actors can exploit these vulnerabilities to propagate toxic content in an automated (or semi-) manner, leveraging the susceptibility of LVLMs to deception via strategically crafted prompts without fine-tuning or compute-intensive procedures. Despite the red-teaming efforts and inherent potential risks associated with the LVLMs, exploring vulnerabilities of LVLMs remains nascent and yet to be fully addressed in a systematic manner. This study systematically examines the vulnerabilities of open-source LVLMs, including LLaVA, InstructBLIP, Fuyu, and Qwen, using adversarial prompt strategies that simulate real-world social manipulation tactics informed by social theories. Our findings show that (i) toxicity and insulting are the most prevalent behaviors, with the mean rates of 16.13% and 9.75%, respectively; (ii) Qwen-VL-Chat, LLaVA-v1.6-Vicuna-7b, and InstructBLIP-Vicuna-7b are the most vulnerable models, exhibiting toxic response rates of 21.50%, 18.30% and 17.90%, and insulting responses of 13.40%, 11.70% and 10.10%, respectively; (iii) prompting strategies incorporating dark humor and multimodal toxic prompt completion significantly elevated these vulnerabilities. Despite being fine-tuned for safety, these models still generate content with varying degrees of toxicity when prompted with adversarial inputs, highlighting the urgent need for enhanced safety mechanisms and robust guardrails in LVLM development.

In this position paper, we review the eclectic recent history of academic and artistic works involving computational systems for humor generation, and focus specifically on live performance. We make the case that AI comedy should be evaluated in live conditions, in front of audiences sharing either physical or online spaces, and under real-time constraints. We further suggest that improvised comedy is therefore the perfect substrate for deploying and assessing computational humor systems. Using examples of successful AI-infused shows, we demonstrate that live performance raises three sets of challenges for computational humor generation: 1) questions around robotic embodiment, anthropomorphism and competition between humans and machines, 2) questions around comedic timing and the nature of audience interaction, and 3) questions about the human interpretation of seemingly absurd AI-generated humor. We argue that these questions impact the choice of methodologies for evaluating computational humor, as any such method needs to work around the constraints of live audiences and performance spaces. These interrogations also highlight different types of collaborative relationship of human comedians towards AI tools.

We describe the construction of a publicly available Yiddish OCR Corpus, and describe and evaluate the open source OCR tool suite Jochre 3, including an Alto editor for corpus annotation, OCR software for Alto OCR layer generation, and a customizable OCR search engine. The current version of the Yiddish OCR corpus contains 658 pages, 186K tokens and 840K glyphs. The Jochre 3 OCR tool uses various fine-tuned YOLOv8 models for top-down page layout analysis, and a custom CNN network for glyph recognition. It attains a CER of 1.5% on our test corpus, far out-performing all other existing public models for Yiddish. We analyzed the full 660M word Yiddish Book Center with Jochre 3 OCR, and the new OCR is searchable through the Yiddish Book Center OCR search engine.

Current evaluations of defenses against prompt attacks in large language model (LLM) applications often overlook two critical factors: the dynamic nature of adversarial behavior and the usability penalties imposed on legitimate users by restrictive defenses. We propose D-SEC (Dynamic Security Utility Threat Model), which explicitly separates attackers from legitimate users, models multi-step interactions, and rigorously expresses the security-utility in an optimizable form. We further address the shortcomings in existing evaluations by introducing Gandalf, a crowd-sourced, gamified red-teaming platform designed to generate realistic, adaptive attack datasets. Using Gandalf, we collect and release a dataset of 279k prompt attacks. Complemented by benign user data, our analysis reveals the interplay between security and utility, showing that defenses integrated in the LLM (e.g., system prompts) can degrade usability even without blocking requests. We demonstrate that restricted application domains, defense-in-depth, and adaptive defenses are effective strategies for building secure and useful LLM applications. Code is available at \href{https://github.com/lakeraai/dsec-gandalf}{\texttt{https://github.com/lakeraai/dsec-gandalf}}.

Long-range sequence modeling is a crucial aspect of natural language processing and time series analysis. However, traditional models like Recurrent Neural Networks (RNNs) and Transformers suffer from computational and memory inefficiencies, especially when dealing with long sequences. This paper introduces Logarithmic Memory Networks (LMNs), a novel architecture that leverages a hierarchical logarithmic tree structure to efficiently store and retrieve past information. LMNs dynamically summarize historical context, significantly reducing the memory footprint and computational complexity of attention mechanisms from O(n2) to O(log(n)). The model employs a single-vector, targeted attention mechanism to access stored information, and the memory block construction worker (summarizer) layer operates in two modes: a parallel execution mode during training for efficient processing of hierarchical tree structures and a sequential execution mode during inference, which acts as a memory management system. It also implicitly encodes positional information, eliminating the need for explicit positional encodings. These features make LMNs a robust and scalable solution for processing long-range sequences in resource-constrained environments, offering practical improvements in efficiency and scalability. The code is publicly available under the MIT License on GitHub: https://github.com/AhmedBoin/LogarithmicMemory.

Fox News correspondent William La Jeunesse joins'Fox News Sunday' to discuss the evolution of AI and the push lawmakers are making to regulate it. FIRST ON FOX: A select group of tech industry titans and venture capitalists will gather in Washington, D.C., this week to welcome the incoming Trump administration and celebrate new opportunities for global innovation in artificial intelligence and entrepreneurship. Presidents and CEOs from companies on the cutting edge of AI tech and their big financial backers, along with personnel from the incoming administration, will attend a dinner on Thursday organized by Outside the Box Ventures, a firm founded last year by journalist-turned-investment banker Katherine Tarbox, along with Laurent Bili, the French ambassador to the U.S. The list of those invited to Thursday's dinner includes "DOGE" chief Elon Musk, Silicon Valley investor and GOP mega-donor Peter Thiel, NVCA chief executive Bobby Franklin, incoming White House AI and crypto czar David Sacks, OpenAI's Sam Altman, investor Joe Lonsdale and Narya co-founder Colin Greenspon. "This gathering represents more than discussion. We hope it symbolizes a new chapter in public-private collaboration to harness technology's transformative power for the nation's future," a source close to the planning told Fox News Digital.