arXiv subjects

Khouloud Saadi

Publications and source records attributed to Khouloud Saadi.

4 recordsLinked to original sources

arXiv2026

Validity-Calibrated Reasoning Distillation

Reasoning distillation aims to transfer multi-step reasoning capabilities from large language models to smaller, more efficient ones. While recent methods have shown promising gains, they typically rely on static teacher-student hierarchies and frame distillation as trajectory imitation. This is misaligned with the structure of reasoning, where intermediate steps are often locally under-specified: global correctness constrains the final answer, but does not uniquely determine each intermediate move. We propose validity-calibrated reasoning distillation, a framework that treats reasoning distillation as a problem of local learning-signal allocation rather than path alignment. Instead of enforcing token-level imitation, we compare the student's and teacher's proposed next-step actions under the same prefix and use their relative local validity to modulate the strength of the distillation update. This yields a dynamic, context-dependent supervision mechanism that preserves the teacher's structural guidance while adapting update strength to local reasoning quality. Across mathematical reasoning, code generation, and instruction-following benchmarks, our method consistently outperforms strong distillation baselines. These results indicate that effective LLM reasoning distillation is governed not by rigid trajectory imitation, but by principled, locally calibrated allocation of learning signal.

cs.LG↗

arXiv2026

What Should Feature Distillation Transfer in LLMs? A Task-Tangent Geometry View

Feature-based knowledge distillation aims to transfer intermediate representations from a teacher LLM model to a student. Existing approaches typically rely on direct feature matching or learned projections, implicitly treating representations as objects with intrinsic meaning. However, the relevance of a representation dimension is determined solely by how it affects the model's output. In this work, we propose a functional perspective on feature-based distillation. We characterize knowledge transfer in terms of the teacher's functional geometry, i.e., how its output depends on internal representations, rather than direct representation alignment. This viewpoint reveals that effective distillation need not preserve full high-dimensional features, but instead should retain dominant directions of functional contribution, naturally inducing an effective functional dimension for each task. Building on this framework, we introduce Flex-KD, an architecture-agnostic and parameter-free distillation method that transfers the teacher's functional geometry while matching the student's representational capacity. Extensive experiments across language understanding and generation benchmarks demonstrate that Flex-KD consistently outperforms existing distillation approaches, particularly under severe teacher-student dimension mismatch.

cs.CL↗

arXiv2025

Dissecting Representation Misalignment in Contrastive Learning via Influence Function

Contrastive learning, commonly applied in large-scale multimodal models, often relies on data from diverse and often unreliable sources, which can include misaligned or mislabeled text-image pairs. This frequently leads to robustness issues and hallucinations, ultimately causing performance degradation. Data valuation is an efficient way to detect and trace these misalignments. Nevertheless, existing methods are computationally expensive for large-scale models. Although computationally efficient, classical influence functions are inadequate for contrastive learning models, as they were initially designed for pointwise loss. Furthermore, contrastive learning involves minimizing the distance between positive sample modalities while maximizing the distance between negative sample modalities. This necessitates evaluating the influence of samples from both perspectives. To tackle these challenges, we introduce the Extended Influence Function for Contrastive Loss (ECIF), an influence function crafted for contrastive loss. ECIF considers both positive and negative samples and provides a closed-form approximation of contrastive learning models, eliminating the need for retraining. Building upon ECIF, we develop a series of algorithms for data evaluation, misalignment detection, and misprediction trace-back tasks. Experimental results demonstrate our ECIF advances the transparency and interpretability of CLIP-style embedding models by offering a more accurate assessment of data impact and model alignment compared to traditional baseline methods.

cs.LG↗

arXiv2021

Effective prevention of semantic drift as angular distance in memory-less continual deep neural networks

Lifelong machine learning or continual learning models attempt to learn incrementally by accumulating knowledge across a sequence of tasks. Therefore, these models learn better and faster. They are used in various intelligent systems that have to interact with humans or any dynamic environment e.g., chatbots and self-driving cars. Memory-less approach is more often used with deep neural networks that accommodates incoming information from tasks within its architecture. It allows them to perform well on all the seen tasks. These models suffer from semantic drift or the plasticity-stability dilemma. The existing models use Minkowski distance measures to decide which nodes to freeze, update or duplicate. These distance metrics do not provide better separation of nodes as they are susceptible to high dimensional sparse vectors. In our proposed approach, we use angular distance to evaluate the semantic drift in individual nodes that provide better separation of nodes and thus better balancing between stability and plasticity. The proposed approach outperforms state-of-the art models by maintaining higher accuracy on standard datasets.

cs.LG↗