arXiv subjects

Robin A. A. Ince

Publications and source records attributed to Robin A. A. Ince.

12 recordsLinked to original sources

arXiv2026

Edu-QuRating: Multi-Dimensional Educational Data Curation with Distilled Pairwise Judgements

Educational data filters have become a practical way to improve language-model pre-training, but most filters treat educational value as a single scalar property. This may be too broad for some applications, especially if the data set already features a high density of educational material. Useful learning material needs to be accurate, engaging, well structured, and appropriate for the intended audience and application (e.g. learner- vs teacher-facing). Following QuRating (Wettig et al. 2024), we introduce Edu-QuRating: a pipeline for multi-dimensional educational data scoring and curation. Edu-QuRating defines education-specific rubrics, uses an LLM judge to label sampled document pairs and distills those pairwise preferences into reusable Edu-QuRaters, which can score individual text chunks on a set of educational criteria. Across two sequence-classification base models and six educational criteria, the best Edu-QuRater recovers held-out GPT-4.1-mini pairwise judgements with mean accuracy 0.917. We then apply the resulting scorers in two applications. First, we investigate the potential of Edu-QuRaters for corpus filtering to improve pretraining of small language models. We scored 322.25M FineWeb-Edu-Fortified documents to obtain a filtered pre-training mixture. In matched single-run pre-training comparisons, models trained with Edu-QuRating-based mixtures reached higher observed aggregate accuracy across nine benchmarks than the FineWeb-Edu baseline, with gains concentrated in particular tasks. Second, we used Edu-QuRater scores as reward terms for GRPO post-training. In held-out pairwise judge evaluations, combining Edu-QuRater and answer-structure rewards produced responses preferred to the Qwen3-4B base model on both pedagogical quality and instruction following.

cs.CL↗

arXiv2026

Visual Reasoning Benchmark: Evaluating Multimodal LLMs on Classroom-Authentic Visual Problems from Primary Education

AI models have achieved state-of-the-art results in textual reasoning; however, their ability to reason over spatial and relational structures remains a critical bottleneck -- particularly in early-grade maths, which relies heavily on visuals. This paper introduces the visual reasoning benchmark (VRB), a novel dataset designed to evaluate Multimodal Large Language Models (MLLMs) on their ability to solve authentic visual problems from classrooms. This benchmark is built on a set of 701 questions sourced from primary school examinations in Zambia and India, which cover a range of tasks such as reasoning by analogy, pattern completion, and spatial matching. We outline the methodology and development of the benchmark which intentionally uses unedited, minimal-text images to test if models can meet realistic needs of primary education. Our findings reveal a ``jagged frontier'' of capability where models demonstrate better proficiency in static skills such as counting and scaling, but reach a distinct ``spatial ceiling'' when faced with dynamic operations like folding, reflection, and rotation. These weaknesses pose a risk for classroom use on visual reasoning problems, with the potential for incorrect marking, false scaffolding, and reinforcing student misconceptions. Consequently, education-focused benchmarks like the VRB are essential for determining the functional boundaries of multimodal tools used in classrooms.

cs.CL↗

arXiv2025

Benchmarking the Pedagogical Knowledge of Large Language Models

Benchmarks like Massive Multitask Language Understanding (MMLU) have played a pivotal role in evaluating AI's knowledge and abilities across diverse domains. However, existing benchmarks predominantly focus on content knowledge, leaving a critical gap in assessing models' understanding of pedagogy - the method and practice of teaching. This paper introduces The Pedagogy Benchmark, a novel dataset designed to evaluate large language models on their Cross-Domain Pedagogical Knowledge (CDPK) and Special Education Needs and Disability (SEND) pedagogical knowledge. These benchmarks are built on a carefully curated set of questions sourced from professional development exams for teachers, which cover a range of pedagogical subdomains such as teaching strategies and assessment methods. Here we outline the methodology and development of these benchmarks. We report results for 97 models, with accuracies spanning a range from 28% to 89% on the pedagogical knowledge questions. We consider the relationship between cost and accuracy and chart the progression of the Pareto value frontier over time. We provide online leaderboards at https://rebrand.ly/pedagogy which are updated with new models and allow interactive exploration and filtering based on various model properties, such as cost per token and open-vs-closed weights, as well as looking at performance in different subjects. LLMs and generative AI have tremendous potential to influence education and help to address the global learning crisis. Education-focused benchmarks are crucial to measure models' capacities to understand pedagogical concepts, respond appropriately to learners' needs, and support effective teaching practices across diverse contexts. They are needed for informing the responsible and evidence-based deployment of LLMs and LLM-based tools in educational settings, and for guiding both development and policy decisions.

cs.CL↗

arXiv2022

A whitening approach for Transfer Entropy permits the application to narrow-band signals

Transfer Entropy, a generalisation of Granger Causality, promises to measure "information transfer" from a source to a target signal by ignoring self-predictability of a target signal when quantifying the source-target relationship. A simple example for signals with such self-predictability are narrowband signals. These are both thought to be intrinsically generated by the brain as well as commonly dealt with in analyses of brain signals, where band-pass filters are used to separate responses from noise. However, the use of Transfer Entropy is usually discouraged in such cases. We simulate simplistic examples where we confirm the failure of classic implementations of Transfer Entropy when applied to narrow-band signals, as made evident by a flawed recovery of effect sizes and interaction delays. We propose an alternative approach based on a whitening of the input signals before computing a bivariate measure of directional time-lagged dependency. This approach solves the problems found in the simple simulated systems. Finally, we explore the behaviour of our measure when applied to delta and theta response components in Magnetoencephalography (MEG) responses to continuous speech. The small effects that our measure attributes to a directed interaction from the stimulus to the neuronal responses are stronger in the theta than in the delta band. This suggests that the delta band reflects a more predictive coupling, while the theta band is stronger involved in bottom-up, reactive processing. Taken together, we hope to increase the interest in directed perspectives on frequency-specific dependencies.

q-bio.NC↗

arXiv2018

Deeper Interpretability of Deep Networks

Deep Convolutional Neural Networks (CNNs) have been one of the most influential recent developments in computer vision, particularly for categorization. There is an increasing demand for explainable AI as these systems are deployed in the real world. However, understanding the information represented and processed in CNNs remains in most cases challenging. Within this paper, we explore the use of new information theoretic techniques developed in the field of neuroscience to enable novel understanding of how a CNN represents information. We trained a 10-layer ResNet architecture to identify 2,000 face identities from 26M images generated using a rigorously controlled 3D face rendering model that produced variations of intrinsic (i.e. face morphology, gender, age, expression and ethnicity) and extrinsic factors (i.e. 3D pose, illumination, scale and 2D translation). With our methodology, we demonstrate that unlike human's network overgeneralizes face identities even with extreme changes of face shape, but it is more sensitive to changes of texture. To understand the processing of information underlying these counterintuitive properties, we visualize the features of shape and texture that the network processes to identify faces. Then, we shed a light into the inner workings of the black box and reveal how hidden layers represent these features and whether the representations are invariant to pose. We hope that our methodology will provide an additional valuable tool for interpretability of CNNs.

cs.CV↗

arXiv2018

Exact partial information decompositions for Gaussian systems based on dependency constraints

The Partial Information Decomposition (PID) [arXiv:1004.2515] provides a theoretical framework to characterize and quantify the structure of multivariate information sharing. A new method (Idep) has recently been proposed for computing a two-predictor PID over discrete spaces. [arXiv:1709.06653] A lattice of maximum entropy probability models is constructed based on marginal dependency constraints, and the unique information that a particular predictor has about the target is defined as the minimum increase in joint predictor-target mutual information when that particular predictor-target marginal dependency is constrained. Here, we apply the Idep approach to Gaussian systems, for which the marginally constrained maximum entropy models are Gaussian graphical models. Closed form solutions for the Idep PID are derived for both univariate and multivariate Gaussian systems. Numerical and graphical illustrations are provided, together with practical and theoretical comparisons of the Idep PID with the minimum mutual information PID (Immi). [arXiv:1411.2832] In particular, it is proved that the Immi method generally produces larger estimates of redundancy and synergy than does the Idep method. In discussion of the practical examples, the PIDs are complemented by the use of deviance tests for the comparison of Gaussian graphical models.

cond-mat.stat-mech↗

arXiv2017

Measuring multivariate redundant information with pointwise common change in surprisal

The problem of how to properly quantify redundant information is an open question that has been the subject of much recent research. Redundant information refers to information about a target variable S that is common to two or more predictor variables Xi. It can be thought of as quantifying overlapping information content or similarities in the representation of S between the Xi. We present a new measure of redundancy which measures the common change in surprisal shared between variables at the local or pointwise level. We provide a game-theoretic operational definition of unique information, and use this to derive constraints which are used to obtain a maximum entropy distribution. Redundancy is then calculated from this maximum entropy distribution by counting only those local co-information terms which admit an unambiguous interpretation as redundant information. We show how this redundancy measure can be used within the framework of the Partial Information Decomposition (PID) to give an intuitive decomposition of the multivariate mutual information into redundant, unique and synergistic contributions. We compare our new measure to existing approaches over a range of example systems, including continuous Gaussian variables. Matlab code for the measure is provided, including all considered examples.

cs.IT↗

arXiv2017

The Partial Entropy Decomposition: Decomposing multivariate entropy and mutual information via pointwise common surprisal

Obtaining meaningful quantitative descriptions of the statistical dependence within multivariate systems is a difficult open problem. Recently, the Partial Information Decomposition (PID) was proposed to decompose mutual information (MI) about a target variable into components which are redundant, unique and synergistic within different subsets of predictor variables. Here, we propose to apply the elegant formalism of the PID to multivariate entropy, resulting in a Partial Entropy Decomposition (PED). We implement the PED with an entropy redundancy measure based on pointwise common surprisal; a natural definition which is closely related to the definition of MI. We show how this approach can reveal the dyadic vs triadic generative structure of multivariate systems that are indistinguishable with classical Shannon measures. The entropy perspective also shows that misinformation is synergistic entropy and hence that MI itself includes both redundant and synergistic effects. We show the relationships between the PED and MI in two predictors, and derive two alternative information decompositions which we illustrate on several example systems. This reveals that in entropy terms, univariate predictor MI is not a proper subset of the joint MI, and we suggest this previously unrecognised fact explains in part why obtaining a consistent PID has proven difficult. The PED also allows separate quantification of mechanistic redundancy (related to the function of the system) versus source redundancy (arising from dependencies between inputs); an important distinction which no existing methods can address. The new perspective provided by the PED helps to clarify some of the difficulties encountered with the PID approach and the resulting decompositions provide useful tools for practical data analysis across a wide range of application areas.

cs.IT↗

arXiv2015

Estimating Information-Theoretic Quantities

Information theory is a practical and theoretical framework developed for the study of communication over noisy channels. Its probabilistic basis and capacity to relate statistical structure to function make it ideally suited for studying information flow in the nervous system. It has a number of useful properties: it is a general measure sensitive to any relationship, not only linear effects; it has meaningful units which in many cases allow direct comparison between different experiments; and it can be used to study how much information can be gained by observing neural responses in single trials, rather than in averages over multiple trials. A variety of information theoretic quantities are in common use in neuroscience - (see entry "Summary of Information-Theoretic Quantities"). Estimating these quantities in an accurate and unbiased way from real neurophysiological data frequently presents challenges, which are explained in this entry.

q-bio.NC↗

arXiv2015

Applications of Information Theory to Analysis of Neural Data

Information theory is a practical and theoretical framework developed for the study of communication over noisy channels. Its probabilistic basis and capacity to relate statistical structure to function make it ideally suited for studying information flow in the nervous system. It has a number of useful properties: it is a general measure sensitive to any relationship, not only linear effects; it has meaningful units which in many cases allow direct comparison between different experiments; and it can be used to study how much information can be gained by observing neural responses in single trials, rather than in averages over multiple trials. A variety of information theoretic quantities are commonly used in neuroscience - (see entry "Definitions of Information-Theoretic Quantities"). In this entry we review some applications of information theory in neuroscience to study encoding of information in both single neurons and neuronal populations.

q-bio.NC↗

arXiv2015

Summary of Information Theoretic Quantities

Information theory is a practical and theoretical framework developed for the study of communication over noisy channels. Its probabilistic basis and capacity to relate statistical structure to function make it ideally suited for studying information flow in the nervous system. As a framework it has a number of useful properties: it provides a general measure sensitive to any relationship, not only linear effects; its quantities have meaningful units which in many cases allow direct comparison between different experiments; and it can be used to study how much information can be gained by observing neural responses in single experimental trials, rather than in averages over multiple trials. A variety of information theoretic quantities are in common use in neuroscience - including the Shannon entropy, Kullback-Leibler divergence, and mutual information. In this entry, we introduce and define these quantities. Further details on how these quantities can be estimated in practice are provided in the entry "Estimation of Information-Theoretic Quantities" and examples of application of these techniques in neuroscience can be found in the entry "Applications of Information-Theoretic Quantities in Neuroscience".

q-bio.NC↗

arXiv2012

Open-source software for studying neural codes

In this chapter we first outline some of the popular computing environments used for analysing neural data, followed by a brief discussion of 'software carpentry', basic tools and skills from software engineering that can be of great use to computational scientists. We then introduce the concept of open-source software and explain some of its potential benefits for the academic community before giving a brief directory of some freely available open source software packages that address various aspects of the study of neural codes. While there are many commercial offerings that provide similar functionality, we concentrate here on open source packages, which in addition to being available free of charge, also have the source code available for study and modification.

q-bio.NC↗