cs.LGNov 18, 2026

Learning Trajectories Leave Decodable Traces in Neural Network Weight Space and Internal Representations

Authors: Patrick Krauss, Achim Schilling, Andreas Maier, Thomas Kinfe, Claus Metzner

Organizations: Cognitive Computational Neuroscience Group, Pattern Recognition Lab, Friedrich-Alexander-University Erlangen-Nürnberg (FAU), Erlangen, Germany · Neural Interface Lab, University Hospital Mannheim, University Heidelberg, Mannheim, Germany · Neuroscience Lab, University Hospital Erlangen, Erlangen, Germany · BG Clinik Ludwigshafen, Ludwigshafen am Rhein, Germany

Abstract

Neural networks trained on the same task develop related weight-space structures, yet remain distinguishable after learning. Do these differences merely reflect alternative solutions, or do they preserve information about individual learning histories? We tested whether final weights reveal initialization, task order, and earlier trajectory states. Training reorganized the ensemble: initial pairwise distances did not consistently predict final distances. Nevertheless, all 30 trained networks were correctly matched to their initial weights among ten candidates per dataset. Networks sharing an initialization also developed recognizable final hidden representations across different batch schedules, although direct comparisons with initial activity did not recover identity. Learning order remained readable as well. A linear probe decoded the order of two earlier tasks with 100% accuracy on 80 held-out networks after ten epochs of a common final task. Earlier intermediate weight states were partly predictable, but individual initial weights could not be reliably reconstructed. Final weights thus retain selective historical information rather than a complete trajectory record. Perturbed networks recovered classification performance without returning to their unperturbed training paths. This separation persisted across datasets, intervention times, and network layers. Moreover, 2,157 of 2,160 aligned endpoint pairs showed no positive loss barrier at sampled interpolation points, indicating low-loss connections between distinct solutions rather than convergence to one weight vector. Together, these findings show that a shared task constrains networks to related successful solutions without erasing their individual histories. Final weights reflect not only what a network has learned, but also where learning began and how it unfolded.

Figures & tables

Appendix figures & tables29 assets

Supplementary material from the paper’s appendix.

Appendix

Explore similar work

Sep 29, 2026

Behavioral Convergence Without Representational Convergence: Persistent Training-History Dependence in Neural Networks

Neural networks trained toward the same final objective can reach similar predictive performance while retaining internal representations shaped by earlier training history. We study this effect using controlled sequential-training experiments in which paired convolutional networks start from identical weights, experience reversed task orders, and then receive the same deterministic common-relaxation distribution. Across 20 paired MNIST runs, 16 satisfy a predeclared behavioral-matching criterion, yet their matched representations retain a mean history score of 0.139 (95% bootstrap CI: 0.127-0.153) and approximately 3.1% prediction disagreement. Extending common relaxation to 50,000 optimizer updates does not erase the measured difference: across five paired seeds, the representation-history score remains 0.190 (95% bootstrap CI: 0.161-0.219) at the end of the measured horizon while the mean accuracy gap is only 0.18 percentage points. Fresh linear probes show that, with sufficient labeled data, the two histories retain practically equivalent linearly accessible class information. A same-label rotated-MNIST control reproduces the effect: all five paired seeds reach behavioral matching while retaining a mean representation-history score of 0.162. Finally, a matched-learning-rate ReLU-LeakyReLU control reduces the 50,000-update representation residue by 0.040 on average in all five paired seeds, providing directional evidence that activation-mediated plasticity contributes to the persistence of training-history effects. These results provide protocol-scoped evidence that behavioral convergence need not imply representational convergence and that optimization history can leave measurable internal traces after prolonged common training.
Oct 7, 2026

When does a network's training history predict its future learning better than its current state? Evidence from a response probe and a forecasting screen

Networks that behave alike now can still learn differently when training continues. Work on loss of plasticity and critical periods shows that the path to a state shapes what follows; it does not show whether the path carries information that a measurement of the state itself misses. We ask when the training history of a network predicts its future learning better than its current state. In a main study, small multilayer perceptrons were trained under three history regimes (42 histories), and future learning was measured at four checkpoints by a short probe: a copy of the network trained for 100 updates on a new task. Before the prediction result was read, the protocol checked the probe. It responded monotonically to a function-preserving rescaling of hidden units, repeated measurements agreed (intraclass correlation 0.940, [0.903, 0.997], in the least reliable class, mean of three repeats), and a re-initialisation of units was visible directly after it but not 100 to 200 updates later. A history state of at most four dimensions did not improve on a calibrated model of the current state (gain -21.4%, 90% interval [-91.9, 8.1]; required in advance: 10%). A companion screen on 1,560 synthetic regression runs asked the same question for a target further away, the final error of the run. There, history models forecast better than the current validation error after 12 of up to 240 epochs (compact state 30.3%, [15.8, 39.4], a contextual comparison) and were not distinguishable from it after 48. In both studies the history was informative only while the current state was not yet informative about the target; this reading was formed after the results.
Sep 28, 2026

Early Learning Shapes Later Directions Of Representation Change In Continual Learning

Representations continually change as a network learns new tasks. We ask whether early representational changes naturally form a geometric structure that continues to shape later learning. We identify a low-dimensional subspace of early representation drift, which we call a scaffold, and test whether it is reused across subsequent tasks. Across four pretrained visual encoders and two datasets, later representational changes consistently favor this early-defined subspace over matched random alternatives. This reuse is history-dependent: when networks experience different early tasks but identical later training inputs, each network preferentially reuses the scaffold induced by its own learning history. The same preference appears in individual optimizer updates, even though the network's dominant local response directions shift away from the original scaffold. Finally, constraining motion within the scaffold slows new-task acquisition more than matched random constraints, while effects on old-task retention are less consistent. In summary, these results suggest that early experience leaves a persistent geometric imprint on how neural networks adapt to future tasks.Code is available at https://github.com/YuantaoDeng/latent-scaffold.