19241 papers
The intersection of computer science and artificial intelligence represents one of the most rapidly evolving frontiers in modern research. This field explores how machines learn from data to solve complex problems, from recognizing patterns in images to generating human-like text. While the underlying mathematics can be dense, the potential applications touch nearly every aspect of daily life, reshaping industries and redefining what is computationally possible.
At Gist.Science, we monitor every new preprint in this category as it appears on arXiv, the primary repository for these breakthroughs. Our team processes each submission immediately, offering both accessible plain-language overviews and detailed technical summaries to ensure the research is understandable for everyone, regardless of their background. This dual approach bridges the gap between raw data and public understanding.
Below are the latest papers in the computer science and artificial intelligence category, freshly summarized for your exploration.
This paper introduces DAPA, a training-free, plug-and-play safety enhancement method that leverages knowledge distillation and model fusion to inject alignment signals from well-aligned models into shadow-aligned ones, significantly improving defense success rates against harmful inputs without compromising performance.
This study provides empirical evidence that the release of ChatGPT causally influenced human spoken communication by increasing the frequency of specific words in spontaneous speech, demonstrating that humans internalize and adopt the lexical patterns of large language models, thereby integrating AI into the ongoing process of cultural evolution.
This paper proposes Generalized Fisher-Weighted SVD (GFWSVD), a scalable post-training compression method for large language models that utilizes a Kronecker-factored approximation of the full Fisher information matrix to capture parameter correlations and significantly outperform existing diagonal-based compression techniques.
This paper introduces SOReL and TOReL, a fully offline Bayesian framework that enables reliable hyperparameter tuning and performance estimation for both model-based and model-free reinforcement learning without any online interactions, while also providing theoretical guarantees of minimax-optimal convergence.
This paper introduces Similarity as Reward Alignment (SARA), a robust contrastive framework for preference-based reinforcement learning that learns latent representations to compute rewards as similarities, demonstrating superior stability and noise resilience compared to baselines on continuous control benchmarks.
This paper proposes a Green AI-guided adaptive encoder freezing strategy for federated learning in MRI-to-CT conversion that significantly reduces energy consumption and CO2 emissions by up to 23% while maintaining or improving model performance, thereby promoting equitable and sustainable healthcare AI.
This paper introduces Step-Tagging, a lightweight framework that uses a novel reasoning step taxonomy (ReasonType) to enable real-time monitoring and early stopping of Language Reasoning Models, achieving a 20–50% reduction in token generation while maintaining comparable accuracy across mathematical and non-mathematical benchmarks.
This study demonstrates that while large language models can be equally effective at convincing people to believe or disbelieve conspiracy theories depending on their instructions, implementing accurate information guardrails and leveraging specific model capabilities can significantly mitigate the risk of AI-driven misinformation.
This study presents a YOLO-based deep learning framework integrated with High-Resolution Class Activation Mapping (HiResCAM) to achieve over 96% accuracy in the automated, interpretable identification of Ichneumonoidea wasp families from high-resolution images, thereby addressing the challenges of manual taxonomic identification in biodiversity and biological control programs.
The paper proposes LBA, a sampling-based method that iteratively integrates prior and posterior knowledge to construct an approximate distribution of high-quality adversarial examples, thereby significantly outperforming existing greedy approaches in generating semantically preserved hard-label adversarial texts under low query budgets.