The field of deep learning is rapidly advancing, with a strong focus on improving the interpretability and representation of complex models. Recent research has made significant progress in developing novel methodologies for understanding how deep learning models represent data, including the use of versatile visualization tools and the exploration of causal factors that influence model similarity. These developments have important implications for a range of applications, from food recognition and brain disease detection to music information retrieval and tobacco quality assessment. Notable papers in this area include: The Spotlight Resonance Method, which provides a novel visualization tool for determining the axis alignment of embedded data; Exploring Causes of Representational Similarity in Machine Learning Models, which investigates the causal factors that influence model similarity; and Moonbeam, a transformer-based foundation model for symbolic music that incorporates music-domain inductive biases. Overall, the field is moving towards a deeper understanding of how deep learning models represent and process complex data, with important implications for both theoretical and practical applications.
Advances in Deep Learning Interpretability and Representation
Sources
The Spotlight Resonance Method: Resolving the Alignment of Embedded Activations
An Explorative Analysis of SVM Classifier and ResNet50 Architecture on African Food Classification
Exploring Causes of Representational Similarity in Machine Learning Models
Enhancing Interpretability of Sparse Latent Representations with Class Information
Moonbeam: A MIDI Foundation Model Using Both Absolute and Relative Music Attributes
Refining Neural Activation Patterns for Layer-Level Concept Discovery in Neural Network-Based Receivers
The Representational Alignment between Humans and Language Models is implicitly driven by a Concreteness Effect
Analyzing Hierarchical Structure in Vision Models with Sparse Autoencoders
An Approach Towards Identifying Bangladeshi Leaf Diseases through Transfer Learning and XAI
An Exploratory Approach Towards Investigating and Explaining Vision Transformer and Transfer Learning for Brain Disease Detection
Layer-wise Investigation of Large-Scale Self-Supervised Music Representation Models
Investigating Fine- and Coarse-grained Structural Correspondences Between Deep Neural Networks and Human Object Image Similarity Judgments Using Unsupervised Alignment
InspectionV3: Enhancing Tobacco Quality Assessment with Deep Convolutional Neural Networks for Automated Workshop Management