eduKate Singapore

Category: Technology

Knowledge Editing | How Models Rewrite Facts, Associations and Parametric Memory Without Full Retraining
Representation Steering | How Activation Vectors and Concept Directions Change Model Behaviour
Activation Patching & Causal Tracing | How Interventions Locate Computation Inside Neural Networks
Linear Probing & Diagnostic Classifiers | What Probes Reveal—and What They Can Learn Themselves
Representation Similarity Analysis | How CKA, CCA, SVCCA, RSA and Procrustes Compare Learned Spaces
Sparse Autoencoders for Neural Representation | How Dictionary Learning Extracts Interpretable Features from Dense Activations
Superposition in Neural Representations | How Models Pack More Features Than Dimensions
Contrastive Representation Learning | How Positive Pairs, Negative Samples, Temperature and InfoNCE Shape Embedding Geometry
Energy-Based Representation Learning | How Models Learn Compatibility Landscapes Instead of Normalising Every Possible World
Manifold Representation Learning | How High-Dimensional Data Becomes Structured Low-Dimensional Geometry
Equivariant Representation Learning | How Models Preserve Structure Through Rotation, Translation, Reflection and Symmetry
Disentangled Representation Learning | How Latent Factors Separate—and Why Independence Is Not Free
Object-Centric Representation | How Scenes Become Persistent Objects, Slots, Relations and Composable World Models
Representation Collapse | Why Learned Embeddings Can Lose Diversity—and How Models Prevent It
Causal Representation Learning | How Models Search for Latent Variables That Survive Interventions and Change

Subscribe to
our newsletter