The Knowledge Microscope: Features as Better Analytical Lenses than Neurons
Previous studies primarily utilize MLP neurons as units of analysis for understanding the mechanisms of factual knowledge in Language Models (LMs); however, neurons suffer from polysemanticity, leading to limited knowledge expression and poor interpretability. In this paper, we first conduct preliminary experiments to validate that Sparse Autoencoders (SAE) can effectively decompose neurons into features, which serve as alternative analytical units. With this established, our core findings reveal three key advantages of features over neurons: (1) Features exhibit stronger influence on knowledge expression and superior interpretability. (2) Features demonstrate enhanced monosemanticity, showing distinct activation patterns between related and unrelated facts. (3) Features achieve better privacy protection than neurons, demonstrated through our proposed FeatureEdit method, which significantly outperforms existing neuron-based approaches in erasing privacy-sensitive information from LMs.Code and dataset will be available.
Code (0)
등록된 구현이 없습니다.
Similar Papers 제목 키워드 기반
Deep learning enhanced mobile-phone microscopy
Mobile-phones have facilitated the creation of field-portable, cost-effective imaging and sensing technologies that approach laboratory-grade instrument performance. However, the optical imaging interfaces of mobile-phon…
Deep LearningLearning to Model and Calibrate Optics via a Differentiable Wave Optics Simulator
We present a novel learning-based method to build a differentiable computational model of a real fluorescence microscope. Our model can be used to calibrate a real optical setup directly from data samples and to engineer…
Deep Learning Microscopy
We demonstrate that a deep neural network can significantly improve optical microscopy, enhancing its spatial resolution over a large field-of-view and depth-of-field. After its training, the only input to this network i…
Deep LearningDarSwin: Distortion Aware Radial Swin Transformer
Wide-angle lenses are commonly used in perception tasks requiring a large field of view. Unfortunately, these lenses produce significant distortions, making conventional models that ignore the distortion effects unable t…
Depth EstimationA low-cost, user-friendly rheo-optical compression assay to measure mechanical properties of cell spheroids in standard cell culture plates
The mechanical characterization of cell spheroids, one of the most widely used 3D biology models in vitro, is a hotspot of current research on the role played by the mechanical response of cells and tissues. The techniqu…