Samy Mammeri

SNIP++: Fine-Grained Symbolic-Numerical Alignment for Symbolic Regression

Kazem Meidani

Mathematical expressions and the numerical behavior they produce are two views of the same underlying function, and connecting them is centr… (voir plus)al to scientific discovery. Symbolic Regression (SR) relies on this connection directly: it searches for an expression that reproduces a given behavior. Recent multi-modal models learn this connection by embedding symbolic expressions and their numerical behavior in a shared representation space. We show that this embedding space is only globally aligned: complete expressions correspond to complete behaviors, but the contribution of individual parts of an expression is not represented. This granularity gap leaves the model unable to tell how a local edit to an expression changes its behavior, the central operation in SR. We introduce a compositional alignment method that closes this gap: a structural positional encoding exposes the substructure of an expression to the encoder, and a multi-granularity contrastive objective grounds each subexpression in the behavior it produces before propagating this grounding to the full expression. The resulting representations close much of the modality gap between symbolic and numerical embeddings, reliably distinguish the effects of local edits that the original alignment cannot, and transfer to external SR corpora.

2026-09-24

arXiv (prépublication)

doi.org

arxiv.org

Hidden-State Similarity Predicts Re-Elicitation After Inoculation Prompting

Fine-tuning on narrow harmful tasks can cause emergent misalignment, where models generalize harmful behavior beyond the training distributi… (voir plus)on. Inoculation prompting can reduce this effect by explicitly eliciting the undesired behavior during training, but recent work shows that the behavior can reappear when evaluation prompts contain cues from the training context. We study what makes such prompts effective triggers. We find that textual similarity to the inoculation prompt is an incomplete predictor: prompts are more likely to re-elicit suppressed behavior when they induce activation states similar to those produced by the inoculation context. These findings advance our understanding of how inoculation prompting modulates conditional misalignment, and suggest that activation-space analysis can help identify when suppressed behaviors remain accessible under eval-time prompts.

2026-06-10

ICML.cc/2026/Workshop/Mech_Interp (poster)

openreview.net

When Does Interleaving Prevent Emergent Misalignment?

Chen Sun

Large language models finetuned on narrow harmful tasks are prone to emergent misalignment (EM), where harmful behavior generalizes beyond t… (voir plus)he training distribution. Interleaving benign data during finetuning has been proposed as a mitigation, but recent work disagrees on whether it prevents EM. In this paper, we investigate this disagreement on Qwen-2.5 7B and 32B, and find that no single property of the interleaved data, taken in isolation, accounts for the gap. Instead, much of it traces to the evaluation itself, as the standard EM benchmark is sensitive to the length of the prompts it uses, and lengthening the evaluation prompts substantially shifts measured misalignment across model sizes. We then identify a region in the model's activations that predicts whether a given interleaved set will prevent EM, and show that reformulating benign data to fall within it substantially reduces EM on both 7B and 32B. This suggests that the standard EM benchmark, which relies on short prompts, may misrepresent the effectiveness of proposed mitigations.

2026-06-10

ICML.cc/2026/Workshop/Mech_Interp (poster)

openreview.net

High-order Component Attribution via Kolmogorov-Arnold Networks

Samy Mammeri

Christian Gagné

Component attribution methods provide insight into how parts of deep learning models, such as convolutional filters and attention heads, inf… (voir plus)luence model predictions. Despite their successes, existing attribution approaches typically assume component effects are additive and independent, neglecting complex interactions among components. Capturing these relations between components is crucial for a better mechanistic understanding of these models. In this work, we improve component attribution (COAR) by replacing the linear counterfactual estimator with a Kolmogorov–Arnold Network (KAN) surrogate fitted to example‑wise perturbation–response data. Then, a symbolic approximation of the learned KAN lets us compute mixed partial derivatives that captures and makes explicit high‑order component interactions that linear methods are missing. These symbolic expressions facilitate future integration with formal verification methods, enabling richer counterfactual analyses of internal model behavior. Preliminary results on standard image classification models demonstrate that our approach improves the accuracy of predicted counterfactuals and enable extraction of higher-order component interactions compared to linear attribution methods.

2025-09-28

NeurIPS.cc/2025/Workshop/MechInterp (poster)

openreview.net

Formation d'entrepreneurs scientifiques

Boussole des politiques en IA

Demandes de supervision

Samy Mammeri

Publications

Formation d'entrepreneurs scientifiques

Boussole des politiques en IA

Demandes de supervision

Mots-clés populaires:

Samy Mammeri

Publications