Early layers of convolutional networks are more interchangeable than later layers across different architectures

Entities in the same semantic neighborhood but without a typed relation to this one — candidates for new edges or unrecognized duplicates.

• Early layers of convolutional neural networks converge to oriented Gabor-like filters, shared with biological visual systemsfinding0.795
  Evidence that convergence to similar representations occurs in early layers across artificial and biological systems
• A vision model trained on ImageNet can be aligned with a model trained on Places-365 while maintaining good performance, and early layers are more interchangeable than later layersfinding0.751
  Lenc & Vedaldi result illustrating data independence in representations and layer-wise alignment
• Layer-wise trajectories show early enrichment, mid-layer alignment, and late re-clustering.claim0.744
  Qualitative geometry pattern.
• Different network depths contribute differentially to the model's capacity for handling deceptive patterns, with middle-to-late layers specializing in abstract deception semanticsclaim0.737
  Interpretation of LAT scanning results showing layer-dependent deception detection accuracy
• Single-layer analyses can be misleading because early-layer truth directions may reflect surface features with limited cross-task generalization.claim0.736
  Methodological critique of prior work that fixed a single layer for truth probing.
• early layers (0–5)concept0.735
  Layers with weak anchoring due to generic representations.
• Convolutional Neural Networksmethod0.728
  Biologically-inspired AI architecture cited as a successful example of bioinspiration from visual cortex organization
• There is a growing similarity in how datapoints are represented in different neural network models, spanning different architectures, training objectives, and data modalitiesclaim0.727
  Primary empirical claim of the paper