Seed-Induced Uniqueness in Transformer Models: Subspace Alignment Governs Subliminal Transfer
We analyze subliminal transfer in Transformer models, where a teacher embeds hidden traits that can be linearly decoded by a student without degrading main-task performance. Prior work often attributes transferability to global representational similarity, typically quantified with Centered Kernel Alignment (CKA). Using synthetic corpora with disentangled public and private labels, we distill students under matched and independent random initializations. We find that transfer strength hinges on alignment within a trait-discriminative subspace: same-seed students inherit this alignment and show higher leakage {τ\approx} 0.24, whereas different-seed students -- despite global CKA > 0.9 -- exhibit substantially reduced excess accuracy {τ\approx} 0.12 - 0.13. We formalize this with subspace-level CKA diagnostic and residualized probes, showing that leakage tracks alignment within the trait-discriminative subspace rather than global representational similarity. Security controls (projection penalty, adversarial reversal, right-for-the-wrong-reasons regularization) reduce leakage in same-base models without impairing public-task fidelity. These results establish seed-induced uniqueness as a resilience property and argue for subspace-aware diagnostics for secure multi-model deployments.
Code (0)
등록된 구현이 없습니다.
Similar Papers 제목 키워드 기반
Subspace-Constrained Federated Learning with Low-Rank Adaptation
Federated low-rank adaptation methods are attractive for fine-tuning large models under communication and privacy constraints, but heterogeneous client data can induce geometric misalignment between local low-rank update…
Federated LearningPolymorphism Is Rotation: Operational Mechanistic Interpretability from a Two-Layer Transformer to Pythia-70m
Independently trained transformers compute the same function in residual-stream bases that differ by a uniform random rotation on $\mathrm{SO}(d_{\mathrm{model}})$. We call this phenomenon polymorphism: same function, mu…
BLANT: Basic Local Alignment of Network Topology, Part 1: Seeding local alignments with unambiguous 8-node graphlets
BLAST is a standard tool in bioinformatics for creating local sequence alignments using a "seed-and-extend" approach. Here we introduce an analogous seed-and-extend algorithm that produces local network alignments: BLANT…
SpecificityA Simplex Witness Certificate and Escape Force for Constant Collapse in Variational Autoencoders
We study exact constant collapse in variational autoencoders: the deterministic encoder mean becomes independent of the input. The prior remains the standard Gaussian. Before VAE training, we select a fixed teacher poste…
Feature Identification via the Empirical NTK
We provide evidence that eigenanalysis of the empirical neural tangent kernel (eNTK) can surface feature directions in trained neural networks. Across three increasingly realistic settings -- a 1-layer MLP trained on mod…