arXiv subjects

Michael Kamp

Publications and source records attributed to Michael Kamp.

At least 19 recordsLinked to original sources

arXiv2026

Tracing Generated Samples to Training-Data Clusters in Flow-Matching Models

Understanding which training samples influence a generated image is an important problem in generative modeling. In flow matching, training samples influence the generated image through the velocity field along the generation trajectory. Removing samples to examine their counterfactual influence changes the velocity field, and the resulting effect on the final image depends on how the change propagates through the trajectory. Consequently, local changes in the velocity field do not necessarily predict the final counterfactual effect. This work investigates attribution in flow-matching models through a hybrid analytical--learned approach, and uses it to derive trajectory-based attribution scores at the cluster level. We evaluate these attribution scores using independently retrained leave-one-cluster-out (LOCO) models, and compare with several attribution baselines using two different flow-matching latent spaces. Our experiments show that semantic similarity constitutes a strong baseline, while the closed-form trajectory-based attribution is competitive in some metrics without requiring counterfactual retraining or model gradients. Our results show that attribution in flow matching depends not only on semantic similarity to training samples, but also on the latent representation, trajectory dynamics, and how influence is propagated to the final output.

cs.LG↗

arXiv2026

Beyond Parameter Aggregation: Semantic Consensus for Federated Fine-Tuning of LLMs

Federated fine-tuning of large language models is commonly formulated as a parameter aggregation problem. However, even parameter-efficient methods require transmitting large collections of trainable weights, assume aligned architectures, and rely on white-box access to model parameters. As model sizes continue to grow and deployments become increasingly heterogeneous, these assumptions become progressively misaligned with practical constraints. We consider an alternative formulation in which collaboration is mediated through model behavior rather than parameters. Clients fine-tune local models on private data and exchange generated outputs on a shared, public prompt set. The server maps these outputs into a semantic representation space, forms a per-prompt semantic consensus, and returns pseudo-labels for further local fine-tuning. This formulation fundamentally changes the communication scaling of federated LLM fine-tuning. The amount of information exchanged depends only on the public prompt budget and the size of the communicated behaviors, independent of model size. As a consequence, the protocol naturally accommodates heterogeneous architectures and applies directly to open-ended text generation. We present a theoretical analysis and empirical results demonstrating that this approach can match strong federated fine-tuning baselines while substantially reducing communication by orders of magnitude (e.g., analytically by a factor of $1006$ for Llama3.1-405B), as well as reductions in runtime and energy consumption. These results suggest that, for generative foundation models, behavior-level consensus provides a more appropriate abstraction for federated adaptation than parameter aggregation.

cs.LG↗

arXiv2026

Exploring and Exploiting Stability in Latent Flow Matching

In this work, we show that Latent Flow-Matching (LFM) models are robust to different types of perturbations, including data reduction and model capacity shrinkage. We characterize this stability by these models' tendency to generate similar outputs under identical noise seeds. We provide a perspective relating this phenomenon to flow matching theory, which indicates that this stability is inherent to the FM objective. We further exploit this stability to derive practical algorithms for more efficient training and inference. Concretely, first, we show that by training LFM models on significantly reduced datasets, performance is preserved, and in compute-constrained regimes, the model converges faster while maintaining quality. This yields multiple advantages, including savings in the training time due to faster convergence, and alleviating annotation effort when training conditional models. Second, LFM stability under architectural shrinkage gives rise to a two-model coarse-to-fine approach, one using a light-weight architecture for the first phase of the FM trajectory, and one with higher capacity for the second, thereby reducing the inference cost substantially. To determine which samples are informative, we introduce three sample-scoring criteria and evaluate them under standard metrics for generative models. Our results are thoroughly evaluated on multiple datasets, demonstrating the practical advantage of this stability, including data savings and a more than two-fold inference speedup while generating comparable outputs.

cs.LG↗

arXiv2026

The Amazing Stability of Flow Matching

The success of deep generative models in generating high-quality and diverse samples is often attributed to particular architectures and large training datasets. In this paper, we investigate the impact of these factors on the quality and diversity of samples generated by \emph{flow-matching} models. Surprisingly, in our experiments on CelebA-HQ dataset, flow matching remains stable even when pruning 50\% of the dataset. That is, the quality and diversity of generated samples are preserved. Moreover, pruning impacts the latent representation only slightly, that is, samples generated by models trained on the full and pruned dataset map to visually similar outputs for a given seed. We observe similar stability when changing the architecture or training configuration, such that the latent representation is maintained under these changes as well. Our results quantify just how strong this stability can be in practice, and help explain the reliability of flow-matching models under various perturbations.

cs.CV↗

arXiv2025

When Flatness Does (Not) Guarantee Adversarial Robustness

Despite their empirical success, neural networks remain vulnerable to small, adversarial perturbations. A longstanding hypothesis suggests that flat minima, regions of low curvature in the loss landscape, offer increased robustness. While intuitive, this connection has remained largely informal and incomplete. By rigorously formalizing the relationship, we show this intuition is only partially correct: flatness implies local but not global adversarial robustness. To arrive at this result, we first derive a closed-form expression for relative flatness in the penultimate layer, and then show we can use this to constrain the variation of the loss in input space. This allows us to formally analyze the adversarial robustness of the entire network. We then show that to maintain robustness beyond a local neighborhood, the loss needs to curve sharply away from the data manifold. We validate our theoretical predictions empirically across architectures and datasets, uncovering the geometric structure that governs adversarial vulnerability, and linking flatness to model confidence: adversarial examples often lie in large, flat regions where the model is confidently wrong. Our results challenge simplified views of flatness and provide a nuanced understanding of its role in robustness.

cs.LG↗

arXiv2025

Flatness is Necessary, Neural Collapse is Not: Rethinking Generalization via Grokking

Neural collapse, i.e., the emergence of highly symmetric, class-wise clustered representations, is frequently observed in deep networks and is often assumed to reflect or enable generalization. In parallel, flatness of the loss landscape has been theoretically and empirically linked to generalization. Yet, the causal role of either phenomenon remains unclear: Are they prerequisites for generalization, or merely by-products of training dynamics? We disentangle these questions using grokking, a training regime in which memorization precedes generalization, allowing us to temporally separate generalization from training dynamics and we find that while both neural collapse and relative flatness emerge near the onset of generalization, only flatness consistently predicts it. Models encouraged to collapse or prevented from collapsing generalize equally well, whereas models regularized away from flat solutions exhibit delayed generalization, resembling grokking, even in architectures and datasets where it does not typically occur. Furthermore, we show theoretically that neural collapse leads to relative flatness under classical assumptions, explaining their empirical co-occurrence. Our results support the view that relative flatness is a potentially necessary and more fundamental property for generalization, and demonstrate how grokking can serve as a powerful probe for isolating its geometric underpinnings.

cs.LG↗

arXiv2025

Whom to Trust? Adaptive Collaboration in Personalized Federated Learning

Data heterogeneity poses a fundamental challenge in federated learning (FL), especially when clients differ not only in distribution but also in the reliability of their predictions across individual examples. While personalized FL (PFL) aims to address this, we observe that many PFL methods fail to outperform two necessary baselines, local training and centralized training. This suggests that meaningful personalization only emerges in a narrow regime, where global models are insufficient, but collaboration across clients still holds value. Our empirical findings point to two key ingredients for success in this regime: adaptivity in collaboration and fine-grained trust, at the level of individual examples. We show that these properties can be achieved within federated semi-supervised learning, where clients exchange predictions over a shared unlabeled dataset. This enables each client to align with public consensus when it is helpful, and disregard it when it is not, without sharing model parameters or raw data. As a concrete realization of this idea, we develop FEDMOSAIC, a personalized co-training method where clients reweight their loss and their contribution to pseudo-labels based on per-example agreement and confidence. FEDMOSAIC outperforms strong FL and PFL baselines across a range of non-IID settings, and we prove convergence under standard smoothness, bounded-variance, and drift assumptions. In contrast to many of these baselines, it also outperforms local and centralized training. These results clarify when federated personalization can be effective, and how fine-grained, trust-aware collaboration enables it.

cs.LG↗

arXiv2024

Federated Binary Matrix Factorization using Proximal Optimization

Identifying informative components in binary data is an essential task in many research areas, including life sciences, social sciences, and recommendation systems. Boolean matrix factorization (BMF) is a family of methods that performs this task by efficiently factorizing the data. In real-world settings, the data is often distributed across stakeholders and required to stay private, prohibiting the straightforward application of BMF. To adapt BMF to this context, we approach the problem from a federated-learning perspective, while building on a state-of-the-art continuous binary matrix factorization relaxation to BMF that enables efficient gradient-based optimization. We propose to only share the relaxed component matrices, which are aggregated centrally using a proximal operator that regularizes for binary outcomes. We show the convergence of our federated proximal gradient descent algorithm and provide differential privacy guarantees. Our extensive empirical evaluation demonstrates that our algorithm outperforms, in terms of quality and efficacy, federation schemes of state-of-the-art BMF methods on a diverse set of real-world and synthetic data.

cs.LG↗

arXiv2024

Landscaping Linear Mode Connectivity

The presence of linear paths in parameter space between two different network solutions in certain cases, i.e., linear mode connectivity (LMC), has garnered interest from both theoretical and practical fronts. There has been significant research that either practically designs algorithms catered for connecting networks by adjusting for the permutation symmetries as well as some others that more theoretically construct paths through which networks can be connected. Yet, the core reasons for the occurrence of LMC, when in fact it does occur, in the highly non-convex loss landscapes of neural networks are far from clear. In this work, we take a step towards understanding it by providing a model of how the loss landscape needs to behave topographically for LMC (or the lack thereof) to manifest. Concretely, we present a `mountainside and ridge' perspective that helps to neatly tie together different geometric features that can be spotted in the loss landscape along the training runs. We also complement this perspective by providing a theoretical analysis of the barrier height, for which we provide empirical support, and which additionally extends as a faithful predictor of layer-wise LMC. We close with a toy example that provides further intuition on how barriers arise in the first place, all in all, showcasing the larger aim of the work -- to provide a working model of the landscape and its topography for the occurrence of LMC.

cs.LG↗

arXiv2024

The Uncanny Valley: Exploring Adversarial Robustness from a Flatness Perspective

Flatness of the loss surface not only correlates positively with generalization, but is also related to adversarial robustness since perturbations of inputs relate non-linearly to perturbations of weights. In this paper, we empirically analyze the relation between adversarial examples and relative flatness with respect to the parameters of one layer. We observe a peculiar property of adversarial examples in the context of relative flatness: during an iterative first-order white-box attack, the flatness of the loss surface measured around the adversarial example first becomes sharper until the label is flipped, but if we keep the attack running, it runs into a flat uncanny valley where the label remains flipped. In extensive experiments, we observe this phenomenon across various model architectures and datasets, even for adversarially trained models. Our results also extend to large language models (LLMs), but due to the discrete nature of the input space and comparatively weak attacks, adversarial examples rarely reach truly flat regions. Most importantly, this phenomenon shows that flatness alone cannot explain adversarial robustness unless we can also guarantee the behavior of the function around the examples. We, therefore theoretically connect relative flatness to adversarial robustness by bounding the third derivative of the loss surface, underlining the need for flatness in combination with a low global Lipschitz constant for a robust model.

cs.LG↗

arXiv2024

Orthogonal Gradient Boosting for Simpler Additive Rule Ensembles

Gradient boosting of prediction rules is an efficient approach to learn potentially interpretable yet accurate probabilistic models. However, actual interpretability requires to limit the number and size of the generated rules, and existing boosting variants are not designed for this purpose. Though corrective boosting refits all rule weights in each iteration to minimise prediction risk, the included rule conditions tend to be sub-optimal, because commonly used objective functions fail to anticipate this refitting. Here, we address this issue by a new objective function that measures the angle between the risk gradient vector and the projection of the condition output vector onto the orthogonal complement of the already selected conditions. This approach correctly approximate the ideal update of adding the risk gradient itself to the model and favours the inclusion of more general and thus shorter rules. As we demonstrate using a wide range of prediction tasks, this significantly improves the comprehensibility/accuracy trade-off of the fitted ensemble. Additionally, we show how objective values for related rule conditions can be computed incrementally to avoid any substantial computational overhead of the new method.

cs.LG↗

arXiv2023

Little is Enough: Boosting Privacy by Sharing Only Hard Labels in Federated Semi-Supervised Learning

In many critical applications, sensitive data is inherently distributed and cannot be centralized due to privacy concerns. A wide range of federated learning approaches have been proposed to train models locally at each client without sharing their sensitive data, typically by exchanging model parameters, or probabilistic predictions (soft labels) on a public dataset or a combination of both. However, these methods still disclose private information and restrict local models to those that can be trained using gradient-based methods. We propose a federated co-training (FedCT) approach that improves privacy by sharing only definitive (hard) labels on a public unlabeled dataset. Clients use a consensus of these shared labels as pseudo-labels for local training. This federated co-training approach empirically enhances privacy without compromising model quality. In addition, it allows the use of local models that are not suitable for parameter aggregation in traditional federated learning, such as gradient-boosted decision trees, rule ensembles, and random forests. Furthermore, we observe that FedCT performs effectively in federated fine-tuning of large language models, where its pseudo-labeling mechanism is particularly beneficial. Empirical evaluations and theoretical analyses suggest its applicability across a range of federated learning scenarios.

cs.LG↗

arXiv2023

Layer-wise Linear Mode Connectivity

Averaging neural network parameters is an intuitive method for fusing the knowledge of two independent models. It is most prominently used in federated learning. If models are averaged at the end of training, this can only lead to a good performing model if the loss surface of interest is very particular, i.e., the loss in the midpoint between the two models needs to be sufficiently low. This is impossible to guarantee for the non-convex losses of state-of-the-art networks. For averaging models trained on vastly different datasets, it was proposed to average only the parameters of particular layers or combinations of layers, resulting in better performing models. To get a better understanding of the effect of layer-wise averaging, we analyse the performance of the models that result from averaging single layers, or groups of layers. Based on our empirical and theoretical investigation, we introduce a novel notion of the layer-wise linear connectivity, and show that deep networks do not have layer-wise barriers between them.

cs.LG↗

arXiv2023

FAM: Relative Flatness Aware Minimization

Flatness of the loss curve around a model at hand has been shown to empirically correlate with its generalization ability. Optimizing for flatness has been proposed as early as 1994 by Hochreiter and Schmidthuber, and was followed by more recent successful sharpness-aware optimization techniques. Their widespread adoption in practice, though, is dubious because of the lack of theoretically grounded connection between flatness and generalization, in particular in light of the reparameterization curse - certain reparameterizations of a neural network change most flatness measures but do not change generalization. Recent theoretical work suggests that a particular relative flatness measure can be connected to generalization and solves the reparameterization curse. In this paper, we derive a regularizer based on this relative flatness that is easy to compute, fast, efficient, and works with arbitrary loss functions. It requires computing the Hessian only of a single layer of the network, which makes it applicable to large neural networks, and with it avoids an expensive mapping of the loss surface in the vicinity of the model. In an extensive empirical evaluation we show that this relative flatness aware minimization (FAM) improves generalization in a multitude of applications and models, both in finetuning and standard training. We make the code available at github.

cs.LG↗

arXiv2023

Why does my medical AI look at pictures of birds? Exploring the efficacy of transfer learning across domain boundaries

It is an open secret that ImageNet is treated as the panacea of pretraining. Particularly in medical machine learning, models not trained from scratch are often finetuned based on ImageNet-pretrained models. We posit that pretraining on data from the domain of the downstream task should almost always be preferred instead. We leverage RadNet-12M, a dataset containing more than 12 million computed tomography (CT) image slices, to explore the efficacy of self-supervised pretraining on medical and natural images. Our experiments cover intra- and cross-domain transfer scenarios, varying data scales, finetuning vs. linear evaluation, and feature space analysis. We observe that intra-domain transfer compares favorably to cross-domain transfer, achieving comparable or improved performance (0.44% - 2.07% performance increase using RadNet pretraining, depending on the experiment) and demonstrate the existence of a domain boundary-related generalization gap and domain-specific learned features.

cs.CV↗

arXiv2022

Open-Source Skull Reconstruction with MONAI

We present a deep learning-based approach for skull reconstruction for MONAI, which has been pre-trained on the MUG500+ skull dataset. The implementation follows the MONAI contribution guidelines, hence, it can be easily tried out and used, and extended by MONAI users. The primary goal of this paper lies in the investigation of open-sourcing codes and pre-trained deep learning models under the MONAI framework. Nowadays, open-sourcing software, especially (pre-trained) deep learning models, has become increasingly important. Over the years, medical image analysis experienced a tremendous transformation. Over a decade ago, algorithms had to be implemented and optimized with low-level programming languages, like C or C++, to run in a reasonable time on a desktop PC, which was not as powerful as today's computers. Nowadays, users have high-level scripting languages like Python, and frameworks like PyTorch and TensorFlow, along with a sea of public code repositories at hand. As a result, implementations that had thousands of lines of C or C++ code in the past, can now be scripted with a few lines and in addition executed in a fraction of the time. To put this even on a higher level, the Medical Open Network for Artificial Intelligence (MONAI) framework tailors medical imaging research to an even more convenient process, which can boost and push the whole field. The MONAI framework is a freely available, community-supported, open-source and PyTorch-based framework, that also enables to provide research contributions with pre-trained models to others. Codes and pre-trained weights for skull reconstruction are publicly available at: https://github.com/Project-MONAI/research-contributions/tree/master/SkullRec

eess.IV↗

arXiv2021

UniFed: A Unified Framework for Federated Learning on Non-IID Image Features

How to tackle non-iid data is a crucial topic in federated learning. This challenging problem not only affects training process, but also harms performance of clients not participating in training. Existing literature mainly focuses on either side, yet still lacks a unified solution to handle these two types (internal and external) of clients in a joint way. In this work, we propose a unified framework to tackle the non-iid issues for internal and external clients together. Firstly, we propose to use client-specific batch normalization in either internal or external clients to alleviate feature distribution shifts incurred by non-iid data. Then we present theoretical analysis to demonstrate the benefits of client-specific batch normalization. Specifically, we show that our approach promotes convergence speed for federated training and yields lower generalization error bound for external clients. Furthermore, we use causal reasoning to form a causal view to explain the advantages of our framework. At last, we conduct extensive experiments on natural and medical images to evaluate our method, where our method achieves state-of-the-art performance, faster convergence, and shows good compatibility. We also performed comprehensive analytical studies on a real-world medical dataset to demonstrate the effectiveness.

cs.LG↗

arXiv2021

Federated Learning from Small Datasets

Federated learning allows multiple parties to collaboratively train a joint model without sharing local data. This enables applications of machine learning in settings of inherently distributed, undisclosable data such as in the medical domain. In practice, joint training is usually achieved by aggregating local models, for which local training objectives have to be in expectation similar to the joint (global) objective. Often, however, local datasets are so small that local objectives differ greatly from the global objective, resulting in federated learning to fail. We propose a novel approach that intertwines model aggregations with permutations of local models. The permutations expose each local model to a daisy chain of local datasets resulting in more efficient training in data-sparse domains. This enables training on extremely small local datasets, such as patient data across hospitals, while retaining the training efficiency and privacy benefits of federated learning.

cs.LG↗