FHE-Agent: Automating CKKS Configuration for Practical Encrypted Inference via an LLM-Guided Agentic Framework

TimePre has been introduced as a novel framework that enhances Probabilistic Time-Series Forecasting (PTSF) by integrating the efficiency of MLP-based models with the flexibility of Multiple Choice Learning (MCL). This development addresses the challenges of computational expense and training instability that have historically limited the performance of generative models in this field.

A novel knowledge distillation framework has been proposed to address shortcut learning in medical image analysis, particularly in deep learning models that may rely on irrelevant features. This framework utilizes a teacher network fine-tuned on relevant data to guide a student network trained on a larger, biased dataset, aiming to enhance the robustness of predictions in high-stakes medical applications.

Recent research has demonstrated that the impact of context in vanilla transformer models can be represented through token-dependent, rank-1 patches to MLP weights. This study extends this theory to modern Large Language Models (LLMs), providing analytical solutions for Gemma-style transformer blocks and generalizing the findings for multi-layer models.

The emergence of Large Language Model (LLM)-driven multi-agent systems has transformed software development, allowing users with minimal technical skills to create applications through natural language inputs. However, this innovation also raises significant security concerns, particularly through scenarios where malicious users exploit benign agents or vice versa. The introduction of the Implicit Malicious Behavior Injection Attack (IMBIA) highlights these vulnerabilities, with alarming success rates in various frameworks.

KGpipe has been introduced as a framework for generating and evaluating pipelines that integrate diverse data sources into knowledge graphs (KGs). This framework addresses the existing gap in combining various methods for information extraction, data transformation, and entity matching into effective end-to-end solutions.

A new method called NVGS has been proposed to enhance 3D Gaussian Splatting by learning viewpoint-dependent visibility functions for occlusion culling, addressing the limitations posed by the semi-transparent nature of Gaussians. This approach utilizes a shared MLP across instances and integrates neural queries into an instanced software rasterizer, improving rendering efficiency and image quality.

The introduction of CleverDistiller marks a significant advancement in self-supervised cross-modal knowledge distillation, enabling the transfer of features from 2D vision foundation models to 3D LiDAR-based models. This framework utilizes a direct feature similarity loss and a multi-layer perceptron projection head, enhancing the learning of complex semantic dependencies in autonomous driving applications.

A recent study has shown that semantic segmentation networks trained on specific lidar types struggle to generalize to new lidar systems without additional intervention. The research focuses on leveraging vision foundation models (VFMs) to enhance unsupervised domain adaptation for semantic segmentation of lidar point clouds, revealing key architectural insights for improving performance across different domains.