The field of artificial intelligence and deep learning is rapidly evolving, with a focus on improving the interpretability, efficiency, and robustness of models. Recent research has explored the use of Infinite Time Turing Machines to analyze deep learning systems, revealing fundamental limitations in scalability, efficiency, and interpretability. Other studies have introduced novel architectures, such as the Universal State Machine and DeepTraverse, which enable more efficient and interpretable computation. Additionally, there has been a growing interest in developing more robust and secure models, with research on attacking attention mechanisms in foundation models and introducing adaptive pooling methods to mitigate noise. Noteworthy papers include 'Infinite Time Turing Machines and their Applications', which proposes a novel computational paradigm, and 'Attacking Attention of Foundation Models Disrupts Downstream Tasks', which highlights the vulnerability of foundation models to adversarial attacks.
Advancements in Artificial Intelligence and Deep Learning
Sources
Infinite Time Turing Machines and their Applications
Attacking Attention of Foundation Models Disrupts Downstream Tasks
A Neural Network Model of Spatial and Feature-Based Attention
Personalized Interpretability -- Interactive Alignment of Prototypical Parts Networks
Transformative or Conservative? Conservation laws for ResNets and Transformers
MARMOT: Masked Autoencoder for Modeling Transient Imaging
Convergence of Spectral Principal Paths: How Deep Networks Distill Linear Representations from Noisy Inputs
Spiking Neural Models for Decision-Making Tasks with Learning
Robust Noise Attenuation via Adaptive Pooling of Transformer Outputs
DeepTraverse: A Depth-First Search Inspired Network for Algorithmic Visual Understanding
Detec\c{c}\~ao da Psor\'iase Utilizando Vis\~ao Computacional: Uma Abordagem Comparativa Entre CNNs e Vision Transformers
Interpreting learned search: finding a transition model and value function in an RNN that plays Sokoban
Revisiting Transformers with Insights from Image Filtering
PiPViT: Patch-based Visual Interpretable Prototypes for Retinal Image Analysis
Saturation Self-Organizing Map
Preserving Task-Relevant Information Under Linear Concept Removal