Linara Adilova

Linara Adilova, Amr Abourayya, Jianning Li et al.

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.

Maximilian Poretschkin, Anna Schmitz, Maram Akila et al.

Artificial Intelligence (AI) has made impressive progress in recent years and represents a key technology that has a crucial impact on the economy and society. However, it is clear that AI and business models based on it can only reach their full potential if AI applications are developed according to high quality standards and are effectively protected against new AI risks. For instance, AI bears the risk of unfair treatment of individuals when processing personal data e.g., to support credit lending or staff recruitment decisions. The emergence of these new risks is closely linked to the fact that the behavior of AI applications, particularly those based on Machine Learning (ML), is essentially learned from large volumes of data and is not predetermined by fixed programmed rules. Thus, the issue of the trustworthiness of AI applications is crucial and is the subject of numerous major publications by stakeholders in politics, business and society. In addition, there is mutual agreement that the requirements for trustworthy AI, which are often described in an abstract way, must now be made clear and tangible. One challenge to overcome here relates to the fact that the specific quality criteria for an AI application depend heavily on the application context and possible measures to fulfill them in turn depend heavily on the AI technology used. Lastly, practical assessment procedures are needed to evaluate whether specific AI applications have been developed according to adequate quality standards. This AI assessment catalog addresses exactly this point and is intended for two target groups: Firstly, it provides developers with a guideline for systematically making their AI applications trustworthy. Secondly, it guides assessors and auditors on how to examine AI applications for trustworthiness in a structured way.

Linara Adilova, Maksym Andriushchenko, Michael Kamp et al.

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.

Nils Philipp Walter, Linara Adilova, Jilles Vreeken et al.

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.

Ting Han, Linara Adilova, Henning Petzka et al.

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.

Maximilian Weimar, Lukas M. Rachbauer, Ilya Starshynov et al.

The estimation of continuous parameters from measured data plays a central role in many fields of physics. A key tool in understanding and improving such estimation processes is the concept of Fisher information, which quantifies how information about unknown parameters propagates through a physical system and determines the ultimate limits of precision. With Artificial Neural Networks (ANNs) gradually becoming an integral part of many measurement systems, it is essential to understand how they process and transmit parameter-relevant information internally. Here, we present a method to monitor the flow of Fisher information through an ANN performing a parameter estimation task, tracking it from the input to the output layer. We show that optimal estimation performance corresponds to the maximal transmission of Fisher information, and that training beyond this point results in information loss due to overfitting. This provides a model-free stopping criterion for network training-eliminating the need for a separate validation dataset. To demonstrate the practical relevance of our approach, we apply it to a network trained on data from an imaging experiment, highlighting its effectiveness in a realistic physical setting.

Sidak Pal Singh, Linara Adilova, Michael Kamp et al.

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.

Linara Adilova, Elena Schulz, Maram Akila et al.

Data-driven sensor interpretation in autonomous driving can lead to highly implausible predictions as can most of the time be verified with common-sense knowledge. However, learning common knowledge only from data is hard and approaches for knowledge integration are an active research area. We propose to use a partly human-designed, partly learned set of rules to describe relations between objects of a traffic scene on a high level of abstraction. In doing so, we improve and robustify existing deep neural networks consuming low-level sensor information. We present an initial study adapting the well-established Probabilistic Soft Logic (PSL) framework to validate and improve on the problem of semantic segmentation. We describe in detail how we integrate common knowledge into the segmentation pipeline using PSL and verify our approach in a set of experiments demonstrating the increase in robustness against several severe image distortions applied to the A2D2 autonomous driving data set.

Linara Adilova, Siming Chen, Michael Kamp

Novelty detection in discrete sequences is a challenging task, since deviations from the process generating the normal data are often small or intentionally hidden. Novelties can be detected by modeling normal sequences and measuring the deviations of a new sequence from the model predictions. However, in many applications data is generated by several distinct processes so that models trained on all the data tend to over-generalize and novelties remain undetected. We propose to approach this challenge through decomposition: by clustering the data we break down the problem, obtaining simpler modeling task in each cluster which can be modeled more accurately. However, this comes at a trade-off, since the amount of training data per cluster is reduced. This is a particular problem for discrete sequences where state-of-the-art models are data-hungry. The success of this approach thus depends on the quality of the clustering, i.e., whether the individual learning problems are sufficiently simpler than the joint problem. While clustering discrete sequences automatically is a challenging and domain-specific task, it is often easy for human domain experts, given the right tools. In this paper, we adapt a state-of-the-art visual analytics tool for discrete sequence clustering to obtain informed clusters from domain experts and use LSTMs to model each cluster individually. Our extensive empirical evaluation indicates that this informed clustering outperforms automatic ones and that our approach outperforms state-of-the-art novelty detection methods for discrete sequences in three real-world application scenarios. In particular, decomposition outperforms a global model despite less training data on each individual cluster.

Lukas Heppe, Michael Kamp, Linara Adilova et al.

The communication between data-generating devices is partially responsible for a growing portion of the world's power consumption. Thus reducing communication is vital, both, from an economical and an ecological perspective. For machine learning, on-device learning avoids sending raw data, which can reduce communication substantially. Furthermore, not centralizing the data protects privacy-sensitive data. However, most learning algorithms require hardware with high computation power and thus high energy consumption. In contrast, ultra-low-power processors, like FPGAs or micro-controllers, allow for energy-efficient learning of local models. Combined with communication-efficient distributed learning strategies, this reduces the overall energy consumption and enables applications that were yet impossible due to limited energy on local devices. The major challenge is then, that the low-power processors typically only have integer processing capabilities. This paper investigates an approach to communication-efficient on-device learning of integer exponential families that can be executed on low-power processors, is privacy-preserving, and effectively minimizes communication. The empirical evaluation shows that the approach can reach a model quality comparable to a centrally learned regular model with an order of magnitude less communication. Comparing the overall energy consumption, this reduces the required energy for solving the machine learning task by a significant amount.

Henning Petzka, Michael Kamp, Linara Adilova et al.

Flatness of the loss curve is conjectured to be connected to the generalization ability of machine learning models, in particular neural networks. While it has been empirically observed that flatness measures consistently correlate strongly with generalization, it is still an open theoretical problem why and under which circumstances flatness is connected to generalization, in particular in light of reparameterizations that change certain flatness measures but leave generalization unchanged. We investigate the connection between flatness and generalization by relating it to the interpolation from representative data, deriving notions of representativeness, and feature robustness. The notions allow us to rigorously connect flatness and generalization and to identify conditions under which the connection holds. Moreover, they give rise to a novel, but natural relative flatness measure that correlates strongly with generalization, simplifies to ridge regression for ordinary least squares, and solves the reparameterization issue.

Henning Petzka, Linara Adilova, Michael Kamp et al.

The performance of deep neural networks is often attributed to their automated, task-related feature construction. It remains an open question, though, why this leads to solutions with good generalization, even in cases where the number of parameters is larger than the number of samples. Back in the 90s, Hochreiter and Schmidhuber observed that flatness of the loss surface around a local minimum correlates with low generalization error. For several flatness measures, this correlation has been empirically validated. However, it has recently been shown that existing measures of flatness cannot theoretically be related to generalization due to a lack of invariance with respect to reparameterizations. We propose a natural modification of existing flatness measures that results in invariance to reparameterization.

Linara Adilova, Julia Rosenzweig, Michael Kamp

An approach to distributed machine learning is to train models on local datasets and aggregate these models into a single, stronger model. A popular instance of this form of parallelization is federated learning, where the nodes periodically send their local models to a coordinator that aggregates them and redistributes the aggregation back to continue training with it. The most frequently used form of aggregation is averaging the model parameters, e.g., the weights of a neural network. However, due to the non-convexity of the loss surface of neural networks, averaging can lead to detrimental effects and it remains an open question under which conditions averaging is beneficial. In this paper, we study this problem from the perspective of information theory: We measure the mutual information between representation and inputs as well as representation and labels in local models and compare it to the respective information contained in the representation of the averaged model. Our empirical results confirm previous observations about the practical usefulness of averaging for neural networks, even if local dataset distributions vary strongly. Furthermore, we obtain more insights about the impact of the aggregation frequency on the information flow and thus on the success of distributed learning. These insights will be helpful both in improving the current synchronization process and in further understanding the effects of model aggregation.

Linara Adilova, Livin Natious, Siming Chen et al.

One of the main tasks of cybersecurity is recognizing malicious interactions with an arbitrary system. Currently, the logging information from each interaction can be collected in almost unrestricted amounts, but identification of attacks requires a lot of effort and time of security experts. We propose an approach for identifying fraud activity through modeling normal behavior in interactions with a system via machine learning methods, in particular LSTM neural networks. In order to enrich the modeling with system specific knowledge, we propose to use an interactive visual interface that allows security experts to identify semantically meaningful clusters of interactions. These clusters incorporate domain knowledge and lead to more precise behavior modeling via informed machine learning. We evaluate the proposed approach on a dataset containing logs of interactions with an administrative interface of login and security server. Our empirical results indicate that the informed modeling is capable of capturing normal behavior, which can then be used to detect abnormal behavior.

Linara Adilova, Nathalie Paul, Peter Schlicht

It has been shown that injecting noise into the neural network weights during the training process leads to a better generalization of the resulting model. Noise injection in the distributed setup is a straightforward technique and it represents a promising approach to improve the locally trained models. We investigate the effects of noise injection into the neural networks during a decentralized training process. We show both theoretically and empirically that noise injection has no positive effect in expectation on linear models, though. However for non-linear neural networks we empirically show that noise injection substantially improves model quality helping to reach a generalization ability of a local model close to the serial baseline.

Linara Adilova, Sven Giesselbach, Stefan Rüping

Scarcity of labeled data is one of the most frequent problems faced in machine learning. This is particularly true in relation extraction in text mining, where large corpora of texts exists in many application domains, while labeling of text data requires an expert to invest much time to read the documents. Overall, state-of-the art models, like the convolutional neural network used in this paper, achieve great results when trained on large enough amounts of labeled data. However, from a practical point of view the question arises whether this is the most efficient approach when one takes the manual effort of the expert into account. In this paper, we report on an alternative approach where we first construct a relation extraction model using distant supervision, and only later make use of a domain expert to refine the results. Distant supervision provides a mean of labeling data given known relations in a knowledge base, but it suffers from noisy labeling. We introduce an active learning based extension, that allows our neural network to incorporate expert feedback and report on first results on a complex data set.

Michael Kamp, Linara Adilova, Joachim Sicking et al.

We propose an efficient protocol for decentralized training of deep neural networks from distributed data sources. The proposed protocol allows to handle different phases of model training equally well and to quickly adapt to concept drifts. This leads to a reduction of communication by an order of magnitude compared to periodically communicating state-of-the-art approaches. Moreover, we derive a communication bound that scales well with the hardness of the serialized learning problem. The reduction in communication comes at almost no cost, as the predictive performance remains virtually unchanged. Indeed, the proposed protocol retains loss bounds of periodically averaging schemes. An extensive empirical evaluation validates major improvement of the trade-off between model performance and communication which could be beneficial for numerous decentralized learning applications, such as autonomous driving, or voice recognition and image classification on mobile phones.