Refusal-Feature-guided Teacher for Safe Finetuning via Data Filtering and Alignment Distillation

Recently, major AI service providers such as Google and OpenAI have introduced Finetuning-as-a-Service, which enables users to customize Large Language Models (LLMs) for specific downstream tasks using their own data. However, this service is vulnerable to degradation of LLM safety-alignment when user data contains harmful prompts. While some prior works address this issue, fundamentally filtering harmful data from user data remains unexplored. Motivated by our observation that a directional representation reflecting refusal behavior (called the refusal feature) obtained from safety-aligned LLMs can inherently distinguish between harmful and harmless prompts, we propose the Refusal-Feature-guided Teacher (ReFT). Our ReFT model is trained to identify harmful prompts based on the similarity between input prompt features and its refusal feature. During finetuning, the ReFT model serves as a teacher that filters harmful prompts from user data and distills alignment knowledge into the base model. Extensive experiments demonstrate that our ReFT-based finetuning strategy effectively minimizes harmful outputs and enhances finetuning accuracy for user-specific tasks, offering a practical solution for secure and reliable deployment of LLMs in Finetuning-as-a-Service.

PECI-Net: Bolus segmentation from video fluoroscopic swallowing study images using preprocessing ensemble and cascaded inference

Bolus segmentation is crucial for the automated detection of swallowing disorders in videofluoroscopic swallowing studies (VFSS). However, it is difficult for the model to accurately segment a bolus region in a VFSS image because VFSS images are translucent, have low contrast and unclear region boundaries, and lack color information. To overcome these challenges, we propose PECI-Net, a network architecture for VFSS image analysis that combines two novel techniques: the preprocessing ensemble network (PEN) and the cascaded inference network (CIN). PEN enhances the sharpness and contrast of the VFSS image by combining multiple preprocessing algorithms in a learnable way. CIN reduces ambiguity in bolus segmentation by using context from other regions through cascaded inference. Moreover, CIN prevents undesirable side effects from unreliably segmented regions by referring to the context in an asymmetric way. In experiments, PECI-Net exhibited higher performance than four recently developed baseline models, outperforming TernausNet, the best among the baseline models, by 4.54\% and the widely used UNet by 10.83\%. The results of the ablation studies confirm that CIN and PEN are effective in improving bolus segmentation performance.

Time Series Using Exponential Smoothing Cells

Time series analysis is used to understand and predict dynamic processes, including evolving demands in business, weather, markets, and biological rhythms. Exponential smoothing is used in all these domains to obtain simple interpretable models of time series and to forecast future values. Despite its popularity, exponential smoothing fails dramatically in the presence of outliers, large amounts of noise, or when the underlying time series changes. We propose a flexible model for time series analysis, using exponential smoothing cells for overlapping time windows. The approach can detect and remove outliers, denoise data, fill in missing observations, and provide meaningful forecasts in challenging situations. In contrast to classic exponential smoothing, which solves a nonconvex optimization problem over the smoothing parameters and initial state, the proposed approach requires solving a single structured convex optimization problem. Recent developments in efficient convex optimization of large-scale dynamic models make the approach tractable. We illustrate new capabilities using synthetic examples, and then use the approach to analyze and forecast noisy real-world time series. Code for the approach and experiments is publicly available.