arXiv subjects

Florian Seiffarth

Publications and source records attributed to Florian Seiffarth.

5 recordsLinked to original sources

arXiv2026

Disentangling Curriculum Learning in NLP: Towards a Unifying Taxonomy

Despite more than a decade of curriculum learning (CL) research in NLP, the field lacks a principled account of which difficulty function or scheduler to use for a given problem. To understand what has hindered progress towards this account, we propose a fine-grained taxonomy separating difficulty evaluation from training scheduling to enable systematic analysis of CL strategies. For difficulty evaluation, we distinguish attribution source and task dependence, revealing difficulty as a perspectival concept encoding different assumptions about what makes an instance hard to learn. For scheduling, we provide the first formalisation of CL schedulers in terms of expected training contribution, enabling comparison across implementations by introducing retention regimes and monotonicity properties. Applied in a dedicated analysis of CL works in NLP, our taxonomy reveals a systematic incomparability problem: prior works conflate distinct notions of difficulty and scheduling, often pursuing different objectives under the same CL label -- hindering comparison and the accumulation of a coherent evidence base. Beyond diagnosis, the taxonomy supports the design, analysis, and comparison of CL strategies, and motivates evaluation practices that disentangle the sources of observed improvement.

cs.CL↗

arXiv2026

Invariant-Based Weight Sharing for Message Passing

Message-passing neural networks (MPNNs) are a powerful framework for learning representations of graph-structured domains. However, weights in MPNNs act on features only, limiting their ability to capture structural patterns. We introduce a novel structure-aware weight sharing principle that explicitly incorporates information inherent to the graph structure. Weights are indexed directly by user-chosen graph invariants, i.e., functions preserved under node permutations, enabling systematic reuse across structurally equivalent subgraphs. We present ShareGNNs, which instantiate this principle within a simple encoder-decoder architecture, resulting in an MPNN with learnable adjacency and transformer-like connectivity. We show that their expressivity is at least as strong as the discriminative power of the chosen invariants, providing explicit control over the model complexity. Experiments on synthetic and real-world data, as well as subgraph counting tasks, demonstrate consistent improvements over standard MPNNs, competitive expressivity beyond the 1-WL test, and scalability to large datasets.

cs.LG↗

arXiv2024

Rule Based Learning with Dynamic (Graph) Neural Networks

A common problem of classical neural network architectures is that additional information or expert knowledge cannot be naturally integrated into the learning process. To overcome this limitation, we propose a two-step approach consisting of (1) generating rule functions from knowledge and (2) using these rules to define rule based layers -- a new type of dynamic neural network layer. The focus of this work is on the second step, i.e., rule based layers that are designed to dynamically arrange learnable parameters in the weight matrices and bias vectors depending on the input samples. Indeed, we prove that our approach generalizes classical feed-forward layers such as fully connected and convolutional layers by choosing appropriate rules. As a concrete application we present rule based graph neural networks (RuleGNNs) that overcome some limitations of ordinary graph neural networks. Our experiments show that the predictive performance of RuleGNNs is comparable to state-of-the-art graph classifiers using simple rules based on Weisfeiler-Leman labeling and pattern counting. Moreover, we introduce new synthetic benchmark graph datasets to show how to integrate expert knowledge into RuleGNNs making them more powerful than ordinary graph neural networks.

cs.LG↗

arXiv2022

A Fast Heuristic for Computing Geodesic Cores in Large Networks

Motivated by the increasing interest in applications of graph geodesic convexity in machine learning and data mining, we present a heuristic for computing the geodesic convex hull of node sets in networks. It generates a set of almost maximal outerplanar spanning subgraphs for the input graph, computes the geodesic closure in each of these graphs, and regards a node as an element of the convex hull if it belongs to the closed sets for at least a user specified number of outerplanar graphs. Our heuristic algorithm runs in time linear in the number of edges of the input graph, i.e., it is faster with one order of magnitude than the standard algorithm computing the closure exactly. Its performance is evaluated empirically by approximating convexity based core-periphery decomposition of networks. Our experimental results with large real-world networks show that for most networks, the proposed heuristic was able to produce close approximations significantly faster than the standard algorithm computing the exact convex hulls. For example, while our algorithm calculated an approximate core-periphery decomposition in 5 hours or less for networks with more than 20 million edges, the standard algorithm did not terminate within 50 days.

cs.SI↗

arXiv2022

Maximal Closed Set and Half-Space Separations in Finite Closure Systems

Several concept learning problems can be regarded as special cases of half-space separation in abstract closure systems over finite ground sets. For the typical scenario that the closure system is implicitly given via a closure operator, we show that the half-space separation problem is NP-complete. As a first approach to overcome this negative result, we relax the problem to maximal closed set separation, give a generic greedy algorithm solving this problem with a linear number of closure operator calls, and show that this bound is sharp. For a second direction, we consider Kakutani closure systems and prove that they are algorithmically characterized by the greedy algorithm. As a first special case of the general problem setting, we consider Kakutani closure systems over graphs and give a sufficient condition for this kind of closure systems in terms of forbidden graph minors. For a second special case, we then focus on closure systems over finite lattices, give an improved adaptation of the generic greedy algorithm, and present an application concerning subsumption lattices.

cs.AI↗