Stefan Horoi

Leveraging Parameter Space Symmetries for Reasoning Skill Transfer in LLMs

Stefan Horoi

Sangwoo Cho

Supriyo Chakraborty

Shi-Xiong Zhang

Sambit Sahu

Guy Wolf

Genta Indra Winata

2025-09-22

NeurIPS.cc/2025/Workshop/UniReps (publié)

openreview.net

Towards a General GNN Framework for Combinatorial Optimization

Michael Perlmutter

2025-07-29

Proceedings of the Third Learning on Graphs Conference (publié)

doi.org

proceedings.mlr.press

Less is More: Undertraining Experts Improves Model Upcycling

Stefan Horoi

Eugene Belilovsky

Guy Wolf

Gintare Karolina Dziugaite

Modern deep learning is increasingly characterized by the use of open-weight foundation models that can be fine-tuned on specialized dataset… (voir plus)s. This has led to a proliferation of expert models and adapters, often shared via platforms like HuggingFace and AdapterHub. To leverage these resources, numerous model upcycling methods have emerged, enabling the reuse of fine-tuned models in multi-task systems. A natural pipeline has thus formed to harness the benefits of transfer learning and amortize sunk training costs: models are pre-trained on general data, fine-tuned on specific tasks, and then upcycled into more general-purpose systems. A prevailing assumption is that improvements at one stage of this pipeline propagate downstream, leading to gains at subsequent steps. In this work, we challenge that assumption by examining how expert fine-tuning affects model upcycling. We show that long fine-tuning of experts that optimizes for their individual performance leads to degraded merging performance, both for fully fine-tuned and LoRA-adapted models, and to worse downstream results when LoRA adapters are upcycled into MoE layers. We trace this degradation to the memorization of a small set of difficult examples that dominate late fine-tuning steps and are subsequently forgotten during merging. Finally, we demonstrate that a task-dependent aggressive early stopping strategy can significantly improve upcycling performance.

2025-06-16

arXiv (Cornell University) (prépublication)

doi.org

arxiv.org

Non-Uniform Parameter-Wise Model Merging

Albert Manuel Orozco Camacho

Stefan Horoi

Guy Wolf

Eugene Belilovsky

Combining multiple machine learning models has long been a technique for enhancing performance, particularly in distributed settings. Tradit… (voir plus)ional approaches, such as model ensembles, work well, but are expensive in terms of memory and compute. Recently, methods based on averaging model parameters have achieved good results in some settings and have gained popularity. However, merging models initialized differently that do not share a part of their training trajectories can yield worse results than simply using the base models, even after aligning their neurons. In this paper, we introduce a novel approach, Non-uniform Parameter-wise Model Merging, or NP Merge, which merges models by learning the contribution of each parameter to the final model using gradient-based optimization. We empirically demonstrate the effectiveness of our method for merging models of various architectures in multiple settings, outperforming past methods. We also extend NP Merge to handle the merging of multiple models, showcasing its scalability and robustness.

2024-12-14

2024 IEEE International Conference on Big Data (BigData) (publié)

doi.org

arxiv.org

Neural networks with optimized single-neuron adaptation uncover biologically plausible regularization

Neurons in the brain have rich and adaptive input-output properties. Features such as heterogeneous f-I curves and spike frequency adaptatio… (voir plus)n are known to place single neurons in optimal coding regimes when facing changing stimuli. Yet, it is still unclear how brain circuits exploit single-neuron flexibility, and how network-level requirements may have shaped such cellular function. To answer this question, a multi-scaled approach is needed where the computations of single neurons and neural circuits must be considered as a complete system. In this work, we use artificial neural networks to systematically investigate single-neuron input-output adaptive mechanisms, optimized in an end-to-end fashion. Throughout the optimization process, each neuron has the liberty to modify its nonlinear activation function, parametrized to mimic f-I curves of biological neurons, and to learn adaptation strategies to modify activation functions in real-time during a task. We find that such networks show much-improved robustness to noise and changes in input statistics. Importantly, we find that this procedure recovers precise coding strategies found in biological neurons, such as gain scaling and fractional order differentiation/integration. Using tools from dynamical systems theory, we analyze the role of these emergent single-neuron properties and argue that neural diversity and adaptation play an active regularization role, enabling neural circuits to optimally propagate information across time.

2024-12-12

PLOS Computational Biology (publié)

doi.org

Understanding Permutation Based Model Merging with Feature Visualizations

Linear mode connectivity (LMC) has become a topic of great interest in recent years. It has been empirically demonstrated that popular deep … (voir plus)learning models trained from different initializations exhibit linear model connectivity up to permutation. Based on this, several approaches for finding a permutation of the model's features or weights have been proposed leading to several popular methods for model merging. These methods enable the simple averaging of two models to create a new high-performance model. However, besides accuracy, the properties of these models and their relationships to the representations of the models they derive from are poorly understood. In this work, we study the inner mechanisms behind LMC in model merging through the lens of classic feature visualization methods. Focusing on convolutional neural networks (CNNs) we make several observations that shed light on the underlying mechanisms of model merging by permute and average.

2024-10-09

NeurIPS.cc/2024/Workshop/UniReps (accepté)

openreview.net

Harmony in Diversity: Merging Neural Networks with Canonical Correlation Analysis

Stefan Horoi

Albert Manuel Orozco Camacho

Eugene Belilovsky

Guy Wolf

Combining the predictions of multiple trained models through ensembling is generally a good way to improve accuracy by leveraging the differ… (voir plus)ent learned features of the models, however it comes with high computational and storage costs. Model fusion, the act of merging multiple models into one by combining their parameters reduces these costs but doesn't work as well in practice. Indeed, neural network loss landscapes are high-dimensional and non-convex and the minima found through learning are typically separated by high loss barriers. Numerous recent works have been focused on finding permutations matching one network features to the features of a second one, lowering the loss barrier on the linear path between them in parameter space. However, permutations are restrictive since they assume a one-to-one mapping between the different models' neurons exists. We propose a new model merging algorithm, CCA Merge, which is based on Canonical Correlation Analysis and aims to maximize the correlations between linear combinations of the model features. We show that our alignment method leads to better performances than past methods when averaging models trained on the same, or differing data splits. We also extend this analysis into the harder setting where more than 2 models are merged, and we find that CCA Merge works significantly better than past methods. Our code is publicly available at https://github.com/shoroi/align-n-merge

2024-07-07

Proceedings of the 41st International Conference on Machine Learning (publié)

doi.org

proceedings.mlr.press

F66. FROM GENE TO COGNITION: MAPPING THE EFFECTS OF GENOMIC DELETIONS AND DUPLICATIONS ON COGNITIVE ABILITY

Sayeh Kazem

Kuldeep Kumar

Thomas Renne

Martineau Jean-Louis

Zohra Saci

Laura Almasy

David Glahn

Guy Wolf

Guillaume Dumas

Sébastien Jacquemont

2023-09-30

European Neuropsychopharmacology (publié)

doi.org

Reliability of CKA as a Similarity Measure in Deep Learning

MohammadReza Davari

Comparing learned neural representations in neural networks is a challenging but important problem, which has been approached in different w… (voir plus)ays. The Centered Kernel Alignment (CKA) similarity metric, particularly its linear variant, has recently become a popular approach and has been widely used to compare representations of a network's different layers, of architecturally similar networks trained differently, or of models with different architectures trained on the same data. A wide variety of conclusions about similarity and dissimilarity of these various representations have been made using CKA. In this work we present analysis that formally characterizes CKA sensitivity to a large class of simple transformations, which can naturally occur in the context of modern machine learning. This provides a concrete explanation of CKA sensitivity to outliers, which has been observed in past works, and to transformations that preserve the linear separability of the data, an important generalization attribute. We empirically investigate several weaknesses of the CKA similarity metric, demonstrating situations in which it gives unexpected or counter-intuitive results. Finally we study approaches for modifying representations to maintain functional behaviour while changing the CKA value. Our results illustrate that, in many cases, the CKA value can be easily manipulated without substantial changes to the functional behaviour of the models, and call for caution when leveraging activation alignment metrics.

2023-01-31

ICLR.cc/2023/Conference (poster)

doi.org

openreview.net

Goal-driven optimization of single-neuron properties in artiﬁcial networks reveals regularization role of neural diversity and adaptation in the brain

Neurons in the brain have rich and adaptive input-output properties. Features such as diverse f-I curves and spike frequency adaptation are … (voir plus)known to place single neurons in optimal coding regimes when facing changing stimuli. Yet, it is still unclear how brain circuits exploit single neuron ﬂexibility, and how network-level requirements may have shaped such cellular function. To answer this question, a multi-scaled approach is needed where the computations of single neurons and of neural circuits must be considered as a complete system. In this work, we use artiﬁcial neural networks to systematically investigate single neuron input-output adaptive mechanisms, optimized in an end-to-end fashion. Throughout the optimization process, each neuron has the liberty to modify its nonlinear activation function, parametrized to mimic f-I curves of biological neurons, and to learn adaptation strategies to modify activation functions in real-time during a task. We ﬁnd that such networks show much-improved robustness to noise and changes in input statistics. Importantly, we ﬁnd that this procedure recovers precise coding strategies found in biological neurons, such as gain scaling and fractional order differentiation/integration. Using tools from dynamical systems theory, we analyze the role of these emergent single neuron properties and argue that neural diversity and adaptation plays an active regularization role that enables neural circuits to optimally propagate information across time.

2021-12-31

(publié)

www.semanticscholar.org

TRAIL : IA responsable pour les professionnels et les leaders

Fondateur en résidence Mila Ventures

Avantage IA : productivité dans la fonction publique

Stefan Horoi

Publications

TRAIL : IA responsable pour les professionnels et les leaders

Fondateur en résidence Mila Ventures

Avantage IA : productivité dans la fonction publique

Mots-clés populaires:

Stefan Horoi

Publications