Telmo M. Silva Filho

Manuel Ferreira-Junior, Jessica T. S. Reinaldo, Telmo M. Silva Filho et al.

Item response theory aims to estimate respondent's latent skills from their responses in tests composed of items with different levels of difficulty. Several models of item response theory have been proposed for different types of tasks, such as binary or probabilistic responses, response time, multiple responses, among others. In this paper, we propose a new version of $β^3$-IRT, called $β^{4}$-IRT, which uses the gradient descent method to estimate the model parameters. In $β^3$-IRT, abilities and difficulties are bounded, thus we employ link functions in order to turn $β^{4}$-IRT into an unconstrained gradient descent process. The original $β^3$-IRT had a symmetry problem, meaning that, if an item was initialised with a discrimination value with the wrong sign, e.g. negative when the actual discrimination should be positive, the fitting process could be unable to recover the correct discrimination and difficulty values for the item. In order to tackle this limitation, we modelled the discrimination parameter as the product of two new parameters, one corresponding to the sign and the second associated to the magnitude. We also proposed sensible priors for all parameters. We performed experiments to compare $β^{4}$-IRT and $β^3$-IRT regarding parameter recovery and our new version outperformed the original $β^3$-IRT. Finally, we made $β^{4}$-IRT publicly available as a Python package, along with the implementation of $β^3$-IRT used in our experiments.

Alex O. Davies, Nirav S. Ajmeri, Telmo M. Silva Filho

Large graphs are present in a variety of domains, including social networks, civil infrastructure, and the physical sciences to name a few. Graph generation is similarly widespread, with applications in drug discovery, network analysis and synthetic datasets among others. While GNN (Graph Neural Network) models have been applied in these domains their high in-memory costs restrict them to small graphs. Conversely less costly rule-based methods struggle to reproduce complex structures. We propose HIGGS (Hierarchical Generation of Graphs) as a model-agnostic framework of producing large graphs with realistic local structures. HIGGS uses GNN models with conditional generation capabilities to sample graphs in hierarchies of resolution. As a result HIGGS has the capacity to extend the scale of generated graphs from a given GNN model by quadratic order. As a demonstration we implement HIGGS using DiGress, a recent graph-diffusion model, including a novel edge-predictive-diffusion variant edge-DiGress. We use this implementation to generate categorically attributed graphs with tens of thousands of nodes. These HIGGS generated graphs are far larger than any previously produced using GNNs. Despite this jump in scale we demonstrate that the graphs produced by HIGGS are, on the local scale, more realistic than those from the rule-based model BTER.

Alex O. Davies, Riku W. Green, Nirav S. Ajmeri et al.

The principal benefit of unsupervised representation learning is that a pre-trained model can be fine-tuned where data or labels are scarce. Existing approaches for graph representation learning are domain specific, maintaining consistent node and edge features across the pre-training and target datasets. This has precluded transfer to multiple domains. We present Topology Only Pre-Training (ToP), a graph pre-training method based on node and edge feature exclusion. We show positive transfer on evaluation datasets from multiple domains, including domains not present in pre-training data, running directly contrary to assumptions made in contemporary works. On 75% of experiments, ToP models perform significantly $p \leq 0.01$ better than a supervised baseline. Performance is significantly positive on 85.7% of tasks when node and edge features are used in fine-tuning. We further show that out-of-domain topologies can produce more useful pre-training than in-domain. Under ToP we show better transfer from non-molecule pre-training, compared to molecule pre-training, on 79% of molecular benchmarks. Against the limited set of other generalist graph models ToP performs strongly, including against models with many orders of magnitude larger. These findings show that ToP opens broad areas of research in both transfer learning on scarcely populated graph domains and in graph foundation models.

Riku Green, Grant Stevens, Zahraa Abdallah et al.

A key aspect of temporal domains is the ability to make predictions multiple time steps into the future, a process known as multi-step forecasting (MSF). At the core of this process is selecting a forecasting strategy, however, with no existing frameworks to map out the space of strategies, practitioners are left with ad-hoc methods for strategy selection. In this work, we propose Stratify, a parameterised framework that addresses multi-step forecasting, unifying existing strategies and introducing novel, improved strategies. We evaluate Stratify on 18 benchmark datasets, five function classes, and short to long forecast horizons (10, 20, 40, 80). In over 84% of 1080 experiments, novel strategies in Stratify improved performance compared to all existing ones. Importantly, we find that no single strategy consistently outperforms others in all task settings, highlighting the need for practitioners explore the Stratify space to carefully search and select forecasting strategies based on task-specific requirements. Our results are the most comprehensive benchmarking of known and novel forecasting strategies. We make code available to reproduce our results.

Riku Green, Huw Day, Zahraa S. Abdallah et al.

Multi-step forecasting is often described through a simple rule of thumb: recursive strategies are said to have high bias and low variance, while direct strategies are said to have low bias and high variance. We revisit this belief by decomposing the expected multi-step forecast error into three parts: irreducible noise, a structural approximation gap, and an estimation-variance term. For linear predictors we show that the structural gap is identically zero for any dataset. For nonlinear predictors, however, the repeated composition used in recursion can increase model expressivity, making the structural gap depend on both the model and the data. We further show that the estimation variance of the recursive strategy at any horizon can be written as the one-step variance multiplied by a Jacobian-based amplification factor that measures how sensitive the composed predictor is to parameter error. This perspective explains when recursive forecasting may simultaneously have lower bias and higher variance than direct forecasting. Experiments with multilayer perceptrons on the ETTm1 dataset confirm these findings. The results offer practical guidance for choosing between recursive and direct strategies based on model nonlinearity and noise characteristics, rather than relying on traditional bias-variance intuition.

Riku Green, Grant Stevens, Telmo de Menezes e Silva Filho et al.

Multi-step forecasting (MSF) in time-series, the ability to make predictions multiple time steps into the future, is fundamental to almost all temporal domains. To make such forecasts, one must assume the recursive complexity of the temporal dynamics. Such assumptions are referred to as the forecasting strategy used to train a predictive model. Previous work shows that it is not clear which forecasting strategy is optimal a priori to evaluating on unseen data. Furthermore, current approaches to MSF use a single (fixed) forecasting strategy. In this paper, we characterise the instance-level variance of optimal forecasting strategies and propose Dynamic Strategies (DyStrat) for MSF. We experiment using 10 datasets from different scales, domains, and lengths of multi-step horizons. When using a random-forest-based classifier, DyStrat outperforms the best fixed strategy, which is not knowable a priori, 94% of the time, with an average reduction in mean-squared error of 11%. Our approach typically triples the top-1 accuracy compared to current approaches. Notably, we show DyStrat generalises well for any MSF task.

Gabriel A. B. do Nascimento, Vincent Dong, Guilherme J. Cavalcante et al.

Accurate breast MRI lesion detection is critical for early cancer diagnosis, especially in high-risk populations. We present a classification pipeline that adapts a pretrained foundation model, the Medical Slice Transformer (MST), for breast lesion classification using dynamic contrast-enhanced MRI (DCE-MRI). Leveraging DINOv2-based self-supervised pretraining, MST generates robust per-slice feature embeddings, which are then used to train a Kolmogorov--Arnold Network (KAN) classifier. The KAN provides a flexible and interpretable alternative to conventional convolutional networks by enabling localized nonlinear transformations via adaptive B-spline activations. This enhances the model's ability to differentiate benign from malignant lesions in imbalanced and heterogeneous clinical datasets. Experimental results demonstrate that the MST+KAN pipeline outperforms the baseline MST classifier, achieving AUC = 0.80 \pm 0.02 while preserving interpretability through attention-based heatmaps. Our findings highlight the effectiveness of combining foundation model embeddings with advanced classification strategies for building robust and generalizable breast MRI analysis tools.

Enrique Crespo-Fernandez, Oliver Ray, Telmo de Menezes e Silva Filho et al.

Efficiently navigating complex environments requires agents to internalize the underlying logic of their world, yet standard world modelling methods often struggle with sample inefficiency, lack of transparency, and poor scalability. We propose a framework for constructing symbolic causal world models entirely online by integrating continuous model learning and repair into the agent's decision loop, by leveraging the power of Meta-Interpretive Learning and predicate invention to find semantically meaningful and reusable abstractions, allowing an agent to construct a hierarchy of disentangled, high-quality concepts from its observations. We demonstrate that our lifted inference approach scales to domains with complex relational dynamics, where propositional methods suffer from combinatorial explosion, while achieving sample-efficiency orders of magnitude higher than the established PPO neural-network-based baseline.

Yu Qin, Phillip Sloan, Raul Santos-Rodriguez et al.

Multimodal models are ubiquitous, yet existing explainability methods are often single-modal, architecture-dependent, or too computationally expensive to run at scale. We introduce LEMON (Local Explanations via Modality-aware OptimizatioN), a model-agnostic framework for local explanations of multimodal predictions. LEMON fits a single modality-aware surrogate with group-structured sparsity to produce unified explanations that disentangle modality-level contributions and feature-level attributions. The approach treats the predictor as a black box and is computationally efficient, requiring relatively few forward passes while remaining faithful under repeated perturbations. We evaluate LEMON on vision-language question answering and a clinical prediction task with image, text, and tabular inputs, comparing against representative multimodal baselines. Across backbones, LEMON achieves competitive deletion-based faithfulness while reducing black-box evaluations by 35-67 times and runtime by 2-8 times compared to strong multimodal baselines.

Alex O. Davies, Roussel Nzoyem, Nirav Ajmeri et al.

Recent research has extensively studied how large language models manipulate integers in specific arithmetic tasks, and on a more fundamental level, how they represent numeric values. These previous works have found that language model embeddings can be used to reconstruct the original values, however, they do not evaluate whether language models actually model continuous values as continuous. Using expected properties of the embedding space, including linear reconstruction and principal component analysis, we show that language models not only represent numeric spaces as non-continuous but also introduce significant noise. Using models from three major providers (OpenAI, Google Gemini and Voyage AI), we show that while reconstruction is possible with high fidelity ($R^2 \geq 0.95$), principal components only explain a minor share of variation within the embedding space. This indicates that many components within the embedding space are orthogonal to the simple numeric input space. Further, both linear reconstruction and explained variance suffer with increasing decimal precision, despite the ordinal nature of the input space being fundamentally unchanged. The findings of this work therefore have implications for the many areas where embedding models are used, in-particular where high numerical precision, large magnitudes or mixed-sign values are common.

Alex O. Davies, Nirav S. Ajmeri, Telmo de Menezes e Silva Filho

Graph learning on molecules makes use of information from both the molecular structure and the features attached to that structure. Much work has been conducted on biasing either towards structure or features, with the aim that bias bolsters performance. Identifying which information source a dataset favours, and therefore how to approach learning that dataset, is an open issue. Here we propose Noise-Noise Ratio Difference (NNRD), a quantitative metric for whether there is more useful information in structure or features. By employing iterative noising on features and structure independently, leaving the other intact, NNRD measures the degradation of information in each. We employ NNRD over a range of molecular tasks, and show that it corresponds well to a loss of information, with intuitive results that are more expressive than simple performance aggregates. Our future work will focus on expanding data domains, tasks and types, as well as refining our choice of baseline model.

Telmo Silva Filho, Hao Song, Miquel Perello-Nieto et al.

This paper provides both an introduction to and a detailed overview of the principles and practice of classifier calibration. A well-calibrated classifier correctly quantifies the level of uncertainty or confidence associated with its instance-wise predictions. This is essential for critical applications, optimal decision making, cost-sensitive classification, and for some types of context change. Calibration research has a rich history which predates the birth of machine learning as an academic field by decades. However, a recent increase in the interest on calibration has led to new methods and the extension from binary to the multiclass setting. The space of options and issues to consider is large, and navigating it requires the right set of concepts and tools. We provide both introductory material and up-to-date technical details of the main concepts and methods, including proper scoring rules and other evaluation metrics, visualisation approaches, a comprehensive account of post-hoc calibration methods for binary and multiclass classification, and several advanced topics.

Meelis Kull, Miquel Perello-Nieto, Markus Kängsepp et al.

Class probabilities predicted by most multiclass classifiers are uncalibrated, often tending towards over-confidence. With neural networks, calibration can be improved by temperature scaling, a method to learn a single corrective multiplicative factor for inputs to the last softmax layer. On non-neural models the existing methods apply binary calibration in a pairwise or one-vs-rest fashion. We propose a natively multiclass calibration method applicable to classifiers from any model class, derived from Dirichlet distributions and generalising the beta calibration method from binary classification. It is easily implemented with neural nets since it is equivalent to log-transforming the uncalibrated probabilities, followed by one linear layer and softmax. Experiments demonstrate improved probabilistic predictions according to multiple measures (confidence-ECE, classwise-ECE, log-loss, Brier score) across a wide range of datasets and classifiers. Parameters of the learned Dirichlet calibration map provide insights to the biases in the uncalibrated model.

Yu Chen, Telmo Silva Filho, Ricardo B. C. Prudêncio et al.

Item Response Theory (IRT) aims to assess latent abilities of respondents based on the correctness of their answers in aptitude test items with different difficulty levels. In this paper, we propose the $β^3$-IRT model, which models continuous responses and can generate a much enriched family of Item Characteristic Curve (ICC). In experiments we applied the proposed model to data from an online exam platform, and show our model outperforms a more standard 2PL-ND model on all datasets. Furthermore, we show how to apply $β^3$-IRT to assess the ability of machine learning classifiers. This novel application results in a new metric for evaluating the quality of the classifier's probability estimates, based on the inferred difficulty and discrimination of data instances.