Michael Witbrock

Qiming Bao, Alex Yuxuan Peng, Tim Hartill et al.

Combining deep learning with symbolic logic reasoning aims to capitalize on the success of both fields and is drawing increasing attention. Inspired by DeepLogic, an end-to-end model trained to perform inference on logic programs, we introduce IMA-GloVe-GA, an iterative neural inference network for multi-step reasoning expressed in natural language. In our model, reasoning is performed using an iterative memory neural network based on RNN with a gated attention mechanism. We evaluate IMA-GloVe-GA on three datasets: PARARULES, CONCEPTRULES V1 and CONCEPTRULES V2. Experimental results show DeepLogic with gated attention can achieve higher test accuracy than DeepLogic and other RNN baseline models. Our model achieves better out-of-distribution generalisation than RoBERTa-Large when the rules have been shuffled. Furthermore, to address the issue of unbalanced distribution of reasoning depths in the current multi-step reasoning datasets, we develop PARARULE-Plus, a large dataset with more examples that require deeper reasoning steps. Experimental results show that the addition of PARARULE-Plus can increase the model's performance on examples requiring deeper reasoning depths. The source code and data are available at https://github.com/Strong-AI-Lab/Multi-Step-Deductive-Reasoning-Over-Natural-Language.

Zhenyun Deng, Yonghua Zhu, Yang Chen et al. · gatech

Effective multi-hop question answering (QA) requires reasoning over multiple scattered paragraphs and providing explanations for answers. Most existing approaches cannot provide an interpretable reasoning process to illustrate how these models arrive at an answer. In this paper, we propose a Question Decomposition method based on Abstract Meaning Representation (QDAMR) for multi-hop QA, which achieves interpretable reasoning by decomposing a multi-hop question into simpler sub-questions and answering them in order. Since annotating the decomposition is expensive, we first delegate the complexity of understanding the multi-hop question to an AMR parser. We then achieve the decomposition of a multi-hop question via segmentation of the corresponding AMR graph based on the required reasoning type. Finally, we generate sub-questions using an AMR-to-Text generation model and answer them with an off-the-shelf QA model. Experimental results on HotpotQA demonstrate that our approach is competitive for interpretable reasoning and that the sub-questions generated by QDAMR are well-formed, outperforming existing question-decomposition-based multi-hop QA approaches.

Qiming Bao, Gael Gendron, Alex Yuxuan Peng et al.

Large language models (LLMs), such as LLaMA, Alpaca, Vicuna, GPT-3.5 and GPT-4, have advanced the performance of AI systems on various natural language processing tasks to human-like levels. However, their generalisation and robustness when performing logical reasoning has not been sufficiently assessed. To comprehensively evaluate this ability, we develop three new logical reasoning datasets named "ReClor-plus", "LogiQA-plus" and "LogiQAv2-plus" that extend standard logical reasoning datasets to evaluate the robustness of the LLM's reasoning. For each, we create three subsets: the first with randomly shuffled options, the second with the correct choices replaced by "none of the other options is correct", and the third with a combination of shuffling and substitution. Experiments on these datasets show that these simple augmentations greatly hinder the models' performance. Despite their high performance on the original publicly available datasets, we find that all models perform poorly on these newly constructed datasets. We also demonstrate that introducing task variations into the training set can markedly improve the model's performance on both the original and our developed datasets. Finally, we show that applying logic-driven data augmentation for fine-tuning and prompting can enhance generalisation in both discriminative and generative models, offering a path to improving their robustness for tasks involving logical reasoning. Source code and data are made publicly available at https://github.com/Strong-AI-Lab/Logical-and-abstract-reasoning.

Zhenyun Deng, Yonghua Zhu, Yang Chen et al. · gatech

Multi-hop question answering (QA) requires reasoning over multiple documents to answer a complex question and provide interpretable supporting evidence. However, providing supporting evidence is not enough to demonstrate that a model has performed the desired reasoning to reach the correct answer. Most existing multi-hop QA methods fail to answer a large fraction of sub-questions, even if their parent questions are answered correctly. In this paper, we propose the Prompt-based Conservation Learning (PCL) framework for multi-hop QA, which acquires new knowledge from multi-hop QA tasks while conserving old knowledge learned on single-hop QA tasks, mitigating forgetting. Specifically, we first train a model on existing single-hop QA tasks, and then freeze this model and expand it by allocating additional sub-networks for the multi-hop QA task. Moreover, to condition pre-trained language models to stimulate the kind of reasoning required for specific multi-hop questions, we learn soft prompts for the novel sub-networks to perform type-specific reasoning. Experimental results on the HotpotQA benchmark show that PCL is competitive for multi-hop QA and retains good performance on the corresponding single-hop sub-questions, demonstrating the efficacy of PCL in mitigating knowledge loss by forgetting.

Gaël Gendron, Michael Witbrock, Gillian Dobbie

Deep Learning models have shown success in a large variety of tasks by extracting correlation patterns from high-dimensional data but still struggle when generalizing out of their initial distribution. As causal engines aim to learn mechanisms independent from a data distribution, combining Deep Learning with Causality can have a great impact on the two fields. In this paper, we further motivate this assumption. We perform an extensive overview of the theories and methods for Causality from different perspectives, with an emphasis on Deep Learning and the challenges met by the two domains. We show early attempts to bring the fields together and the possible perspectives for the future. We finish by providing a large variety of applications for techniques from Causality.

Nathan Young, Qiming Bao, Joshua Bensemann et al.

Transformers have recently been shown to be capable of reliably performing logical reasoning over facts and rules expressed in natural language, but abductive reasoning - inference to the best explanation of an unexpected observation - has been underexplored despite significant applications to scientific discovery, common-sense reasoning, and model interpretability. We present AbductionRules, a group of natural language datasets designed to train and test generalisable abduction over natural-language knowledge bases. We use these datasets to finetune pretrained Transformers and discuss their performance, finding that our models learned generalisable abductive techniques but also learned to exploit the structure of our data. Finally, we discuss the viability of this approach to abductive reasoning and ways in which it may be improved in future work.

Qiming Bao, Juho Leinonen, Paul Denny et al.

Direct Preference Optimization (DPO), the efficient alternative to PPO-based RLHF, falls short on knowledge-intensive generation: standard preference signals from human annotators or LLM judges exhibit a systematic verbosity bias that rewards fluency over logical correctness. This blindspot leaves a logical alignment gap -- SFT models reach NLI entailment of only 0.05-0.22 despite producing fluent text. We propose RLearner-LLM with Hybrid-DPO: an automated preference pipeline that fuses a DeBERTa-v3 NLI signal with a verifier LLM score, removing human annotation while overcoming the "alignment tax" of single-signal optimization. Evaluated across five academic domains (Biology, Medicine, Law) with three base architectures (LLaMA-2-13B, Qwen3-8B, Gemma 4 E4B-it), RLearner-LLM yields up to 6x NLI improvement over SFT, with NLI gains in 11 of 15 cells and consistent answer-coverage gains. On Gemma 4 E4B-it (4.5B effective params), Hybrid-DPO lifts NLI in four of five domains (+11.9% to +2.4x) with faster inference across all five, scaling down to compact base models without losing the alignment-tax mitigation. Our Qwen3-8B RLearner-LLM wins 95% of pairwise comparisons against its own SFT baseline; GPT-4o-mini in turn wins 95% against our concise output -- alongside the 69% win the same judge gives a verbose SFT over our DPO model, this replicates verbosity bias on a frontier comparator and motivates logic-aware metrics (NLI, ACR) over LLM-as-a-judge for knowledge-intensive generation.

Mitchell Rogers, Gaël Gendron, David Arturo Soriano Valdez et al. · gatech

Recording animal behaviour is an important step in evaluating the well-being of animals and further understanding the natural world. Current methods for documenting animal behaviour within a zoo setting, such as scan sampling, require excessive human effort, are unfit for around-the-clock monitoring, and may produce human-biased results. Several animal datasets already exist that focus predominantly on wildlife interactions, with some extending to action or behaviour recognition. However, there is limited data in a zoo setting or data focusing on the group behaviours of social animals. We introduce a large meerkat (Suricata Suricatta) behaviour recognition video dataset with diverse annotated behaviours, including group social interactions, tracking of individuals within the camera view, skewed class distribution, and varying illumination conditions. This dataset includes videos from two positions within the meerkat enclosure at the Wellington Zoo (Wellington, New Zealand), with 848,400 annotated frames across 20 videos and 15 unannotated videos.

Qiming Bao, Juho Leinonen, Alex Yuxuan Peng et al.

Large language models exhibit superior capabilities in processing and understanding language, yet their applications in educational contexts remain underexplored. Learnersourcing enhances learning by engaging students in creating their own educational content. When learnersourcing multiple-choice questions, creating explanations for the solution of a question is a crucial step; it helps other students understand the solution and promotes a deeper understanding of related concepts. However, it is often difficult for students to craft effective solution explanations, due to limited subject understanding. To help scaffold the task of automated explanation generation, we present and evaluate a framework called "ILearner-LLM", that iteratively enhances the generated explanations for the given questions with large language models. Comprising an explanation generation model and an explanation evaluation model, the framework generates high-quality student-aligned explanations by iteratively feeding the quality rating score from the evaluation model back into the instruction prompt of the explanation generation model. Experimental results demonstrate the effectiveness of our ILearner-LLM on LLaMA2-13B and GPT-4 to generate higher quality explanations that are closer to those written by students on five PeerWise datasets. Our findings represent a promising path to enrich the learnersourcing experience for students and to enhance the capabilities of large language models for educational applications.

Neşet Özkan Tan, Alex Yuxuan Peng, Joshua Bensemann et al.

Identifying words that impact a task's performance more than others is a challenge in natural language processing. Transformers models have recently addressed this issue by incorporating an attention mechanism that assigns greater attention (i.e., relevance) scores to some words than others. Because of the attention mechanism's high computational cost, transformer models usually have an input-length limitation caused by hardware constraints. This limitation applies to many transformers, including the well-known bidirectional encoder representations of the transformer (BERT) model. In this paper, we examined BERT's attention assignment mechanism, focusing on two questions: (1) How can attention be employed to reduce input length? (2) How can attention be used as a control mechanism for conditional text generation? We investigated these questions in the context of a text classification task. We discovered that BERT's early layers assign more critical attention scores for text classification tasks compared to later layers. We demonstrated that the first layer's attention sums could be used to filter tokens in a given sequence, considerably decreasing the input length while maintaining good test accuracy. We also applied filtering, which uses a compute-efficient semantic similarities algorithm, and discovered that retaining approximately 6\% of the original sequence is sufficient to obtain 86.5\% accuracy. Finally, we showed that we could generate data in a stable manner and indistinguishable from the original one by only using a small percentage (10\%) of the tokens with high attention scores according to BERT's first layer.

Tim Hartill, Neset Tan, Michael Witbrock et al.

We equip a smaller Language Model to generalise to answering challenging compositional questions that have not been seen in training. To do so we propose a combination of multitask supervised pretraining on up to 93 tasks designed to instill diverse reasoning abilities, and a dense retrieval system that aims to retrieve a set of evidential paragraph fragments. Recent progress in question-answering has been achieved either through prompting methods against very large pretrained Language Models in zero or few-shot fashion, or by fine-tuning smaller models, sometimes in conjunction with information retrieval. We focus on the less explored question of the extent to which zero-shot generalisation can be enabled in smaller models with retrieval against a corpus within which sufficient information to answer a particular question may not exist. We establish strong baselines in this setting for diverse evaluation datasets (StrategyQA, CommonsenseQA, IIRC, DROP, Musique and ARC-DA), and show that performance can be significantly improved by adding retrieval-augmented training datasets which are designed to expose our models to a variety of heuristic reasoning strategies such as weighing partial evidence or ignoring an irrelevant context.

Michael Witbrock, Patrick Haffner

We present SVCnet, a system for modelling speaker variability. Encoder Neural Networks specialized for each speech sound produce low dimensionality models of acoustical variation, and these models are further combined into an overall model of voice variability. A training procedure is described which minimizes the dependence of this model on which sounds have been uttered. Using the trained model (SVCnet) and a brief, unconstrained sample of a new speaker's voice, the system produces a Speaker Voice Code that can be used to adapt a recognition system to the new speaker without retraining. A system which combines SVCnet with an MS-TDNN recognizer is described

Tim Hartill, Diana Benavides-Prado, Michael Witbrock et al.

When provided with sufficient explanatory context, smaller Language Models have been shown to exhibit strong reasoning ability on challenging short-answer question-answering tasks where the questions are unseen in training. We evaluate two methods for further improvement in this setting. Both methods focus on combining rationales generated by a larger Language Model with longer contexts created from a multi-hop dense retrieval system. The first method ($\textit{RR}$) involves training a Rationale Ranking model to score both generated rationales and retrieved contexts with respect to relevance and truthfulness. We then use the scores to derive combined contexts from both knowledge sources using a number of combinatory strategies. For the second method ($\textit{RATD}$) we utilise retrieval-augmented training datasets developed by Hartill et al. 2023 to train a smaller Reasoning model such that it becomes proficient at utilising relevant information from longer text sequences that may be only partially evidential and frequently contain many irrelevant sentences. We find that both methods significantly improve results. Our single best Reasoning model materially improves upon strong comparable prior baselines for unseen evaluation datasets (StrategyQA 58.9 $\rightarrow$ 61.7 acc., CommonsenseQA 63.6 $\rightarrow$ 72.7 acc., ARC-DA 31.6 $\rightarrow$ 52.1 F1, IIRC 25.5 $\rightarrow$ 27.3 F1) and a version utilising our prior knowledge of each type of question in selecting a context combination strategy does even better. Our proposed models also generally outperform direct prompts against much larger models (BLOOM 175B and StableVicuna 13B) in both few-shot chain-of-thought and standard few-shot settings.

Gaël Gendron, Michael Witbrock, Gillian Dobbie

The process of generating data such as images is controlled by independent and unknown factors of variation. The retrieval of these variables has been studied extensively in the disentanglement, causal representation learning, and independent component analysis fields. Recently, approaches merging these domains together have shown great success. Instead of directly representing the factors of variation, the problem of disentanglement can be seen as finding the interventions on one image that yield a change to a single factor. Following this assumption, we introduce a new method for disentanglement inspired by causal dynamics that combines causality theory with vector-quantized variational autoencoders. Our model considers the quantized vectors as causal variables and links them in a causal graph. It performs causal interventions on the graph and generates atomic transitions affecting a unique factor of variation in the image. We also introduce a new task of action retrieval that consists of finding the action responsible for the transition between two images. We test our method on standard synthetic and real-world disentanglement datasets. We show that it can effectively disentangle the factors of variation and perform precise interventions on high-level semantic attributes of an image without affecting its quality, even with imbalanced data distributions.

Libo Zhang, Yang Chen, Toru Takisaka et al.

Correlated Equilibrium (CE) is a well-established solution concept that captures coordination among agents and enjoys good algorithmic properties. In real-world multi-agent systems, in addition to being in an equilibrium, agents' policies are often expected to meet requirements with respect to safety, and fairness. Such additional requirements can often be expressed in terms of the state density which measures the state-visitation frequencies during the course of a game. However, existing CE notions or CE-finding approaches cannot explicitly specify a CE with particular properties concerning state density; they do so implicitly by either modifying reward functions or using value functions as the selection criteria. The resulting CE may thus not fully fulfil the state-density requirements. In this paper, we propose Density-Based Correlated Equilibria (DBCE), a new notion of CE that explicitly takes state density as selection criterion. Concretely, we instantiate DBCE by specifying different state-density requirements motivated by real-world applications. To compute DBCE, we put forward the Density Based Correlated Policy Iteration algorithm for the underlying control problem. We perform experiments on various games where results demonstrate the advantage of our CE-finding approach over existing methods in scenarios with state-density concerns.

Stefan Fuchs, Michael Witbrock, Johannes Dimyadi et al.

Compliance checking is an essential part of a construction project. The recent rapid uptake of building information models (BIM) in the construction industry has created more opportunities for automated compliance checking (ACC). BIM enables sharing of digital building design data that can be used for compliance checking with legal requirements, which are conventionally conveyed in natural language and not intended for machine processing. Creating a computable representation of legal requirements suitable for ACC is complex, costly, and time-consuming. Large language models (LLMs) such as the generative pre-trained transformers (GPT), GPT-3.5 and GPT-4, powering OpenAI's ChatGPT, can generate logically coherent text and source code responding to user prompts. This capability could be used to automate the conversion of building regulations into a semantic and computable representation. This paper evaluates the performance of LLMs in translating building regulations into LegalRuleML in a few-shot learning setup. By providing GPT-3.5 with only a few example translations, it can learn the basic structure of the format. Using a system prompt, we further specify the LegalRuleML representation and explore the existence of expert domain knowledge in the model. Such domain knowledge might be ingrained in GPT-3.5 through the broad pre-training but needs to be brought forth by careful contextualisation. Finally, we investigate whether strategies such as chain-of-thought reasoning and self-consistency could apply to this use case. As LLMs become more sophisticated, the increased common sense, logical coherence, and means to domain adaptation can significantly support ACC, leading to more efficient and effective checking processes.

Yuchen Su, Shaoxin Zhong, Yonghua Zhu et al.

Puns represent a typical linguistic phenomenon that exploits polysemy and phonetic ambiguity to generate humour, posing unique challenges for natural language understanding. Within pun research, audio plays a central role in human communication except text and images, while datasets and systematic resources for spoken puns remain scarce, leaving this crucial modality largely underexplored. In this paper, we present APUN-Bench, the first benchmark dedicated to evaluating large audio language models (LALMs) on audio pun understanding. Our benchmark contains 4,434 audio samples annotated across three stages: pun recognition, pun word location and pun meaning inference. We conduct a deep analysis of APUN-Bench by systematically evaluating 10 state-of-the-art LALMs, uncovering substantial performance gaps in recognizing, localizing, and interpreting audio puns. This analysis reveals key challenges, such as positional biases in audio pun location and error cases in meaning inference, offering actionable insights for advancing humour-aware audio intelligence.

Shaoxin Zhong, Yuchen Su, Michael Witbrock

Mitigating elderly loneliness requires policy interventions that achieve both adaptability and auditability. Existing methods struggle to reconcile these objectives: traditional agent-based models suffer from static rigidity, while direct large language model (LLM) controllers lack essential traceability. This work proposes a three-layer framework that separates diagnosis from control to achieve both properties simultaneously. LLMs operate strictly as diagnostic instruments that assess population state and generate structured risk evaluations, while deterministic formulas with explicit bounds translate these assessments into traceable parameter updates. This separation ensures that every policy decision can be attributed to inspectable rules while maintaining adaptive response to emergent needs. We validate the framework through systematic ablation across five experimental conditions in elderly care simulation. Results demonstrate that explicit control rules outperform end-to-end black-box LLM approaches by 11.7\% while preserving full auditability, confirming that transparency need not compromise adaptive performance.

Tim Hartill, Joshua Bensemann, Michael Witbrock et al.

A distinction is often drawn between a model's ability to predict a label for an evaluation sample that is directly memorised from highly similar training samples versus an ability to predict the label via some method of generalisation. In the context of using Language Models for question-answering, discussion continues to occur as to the extent to which questions are answered through memorisation. We consider this issue for questions that would ideally be answered through reasoning over an associated context. We propose a method of identifying evaluation samples for which it is very unlikely our model would have memorised the answers. Our method is based on semantic similarity of input tokens and label tokens between training and evaluation samples. We show that our method offers advantages upon some prior approaches in that it is able to surface evaluation-train pairs that have overlap in either contiguous or discontiguous sequences of tokens. We use this method to identify unmemorisable subsets of our evaluation datasets. We train two Language Models in a multitask fashion whereby the second model differs from the first only in that it has two additional datasets added to the training regime that are designed to impart simple numerical reasoning strategies of a sort known to improve performance on some of our evaluation datasets but not on others. We then show that there is performance improvement between the two models on the unmemorisable subsets of the evaluation datasets that were expected to benefit from the additional training datasets. Specifically, performance on unmemorisable subsets of two of our evaluation datasets, DROP and ROPES significantly improves by 9.0%, and 25.7% respectively while other evaluation datasets have no significant change in performance.

Yuchen Su, Yonghua Zhu, Yang Chen et al.

Humour translation plays a vital role as a bridge between different cultures, fostering understanding and communication. Although most existing Large Language Models (LLMs) are capable of general translation tasks, these models still struggle with humour translation, which is especially reflected through linguistic interference and lacking humour in translated text. In this paper, we propose a psychology-inspired Humour Decomposition Mechanism (HDM) that utilises Chain-of-Thought (CoT) to imitate the ability of the human thought process, stimulating LLMs to optimise the readability of translated humorous texts. Moreover, we integrate humour theory in HDM to further enhance the humorous elements in the translated text. Our automatic evaluation experiments on open-source humour datasets demonstrate that our method significantly improves the quality of humour translation, yielding average gains of 7.75\% in humour, 2.81\% in fluency, and 6.13\% in coherence of the generated text.

Nathan Young, Michael Witbrock

Solomonoff Induction is an optimal-in-the-limit unbounded algorithm for sequence prediction, representing a Bayesian mixture of every computable probability distribution and performing close to optimally in predicting any computable sequence. Being an optimal form of computational sequence prediction, it seems plausible that it may be used as a model against which other methods of sequence prediction might be compared. We put forth and explore the hypothesis that Transformer models - the basis of Large Language Models - approximate Solomonoff Induction better than any other extant sequence prediction method. We explore evidence for and against this hypothesis, give alternate hypotheses that take this evidence into account, and outline next steps for modelling Transformers and other kinds of AI in this way.

Qiming Bao, Alex Yuxuan Peng, Zhenyun Deng et al.

Combining large language models with logical reasoning enhances their capacity to address problems in a robust and reliable manner. Nevertheless, the intricate nature of logical reasoning poses challenges when gathering reliable data from the web to build comprehensive training datasets, subsequently affecting performance on downstream tasks. To address this, we introduce a novel logic-driven data augmentation approach, AMR-LDA. AMR-LDA converts the original text into an Abstract Meaning Representation (AMR) graph, a structured semantic representation that encapsulates the logical structure of the sentence, upon which operations are performed to generate logically modified AMR graphs. The modified AMR graphs are subsequently converted back into text to create augmented data. Notably, our methodology is architecture-agnostic and enhances both generative large language models, such as GPT-3.5 and GPT-4, through prompt augmentation, and discriminative large language models through contrastive learning with logic-driven data augmentation. Empirical evidence underscores the efficacy of our proposed method with improvement in performance across seven downstream tasks, such as reading comprehension requiring logical reasoning, textual entailment, and natural language inference. Furthermore, our method leads on the ReClor leaderboard at https://eval.ai/web/challenges/challenge-page/503/leaderboard/1347. The source code and data are publicly available at https://github.com/Strong-AI-Lab/Logical-Equivalence-driven-AMR-Data-Augmentation-for-Representation-Learning.

Lingfei Wu, Ian En-Hsu Yen, Jinfeng Yi et al.

Time series data analytics has been a problem of substantial interests for decades, and Dynamic Time Warping (DTW) has been the most widely adopted technique to measure dissimilarity between time series. A number of global-alignment kernels have since been proposed in the spirit of DTW to extend its use to kernel-based estimation method such as support vector machine. However, those kernels suffer from diagonal dominance of the Gram matrix and a quadratic complexity w.r.t. the sample size. In this work, we study a family of alignment-aware positive definite (p.d.) kernels, with its feature embedding given by a distribution of \emph{Random Warping Series (RWS)}. The proposed kernel does not suffer from the issue of diagonal dominance while naturally enjoys a \emph{Random Features} (RF) approximation, which reduces the computational complexity of existing DTW-based techniques from quadratic to linear in terms of both the number and the length of time-series. We also study the convergence of the RF approximation for the domain of time series of unbounded length. Our extensive experiments on 16 benchmark datasets demonstrate that RWS outperforms or matches state-of-the-art classification and clustering methods in both accuracy and computational time. Our code and data is available at { \url{https://github.com/IBM/RandomWarpingSeries}}.

Shiyu Chang, Yang Zhang, Wei Han et al.

Learning with recurrent neural networks (RNNs) on long sequences is a notoriously difficult task. There are three major challenges: 1) complex dependencies, 2) vanishing and exploding gradients, and 3) efficient parallelization. In this paper, we introduce a simple yet effective RNN connection structure, the DilatedRNN, which simultaneously tackles all of these challenges. The proposed architecture is characterized by multi-resolution dilated recurrent skip connections and can be combined flexibly with diverse RNN cells. Moreover, the DilatedRNN reduces the number of parameters needed and enhances training efficiency significantly, while matching state-of-the-art performance (even with standard RNN cells) in tasks involving very long-term dependencies. To provide a theory-based quantification of the architecture's advantages, we introduce a memory capacity measure, the mean recurrent length, which is more suitable for RNNs with long skip connections than existing measures. We rigorously prove the advantages of the DilatedRNN over other recurrent neural architectures. The code for our method is publicly available at https://github.com/code-terminator/DilatedRNN

Gaël Gendron, Bao Trung Nguyen, Alex Yuxuan Peng et al.

Despite impressive performance on language modelling and complex reasoning tasks, Large Language Models (LLMs) fall short on the same tasks in uncommon settings or with distribution shifts, exhibiting a lack of generalisation ability. By contrast, systems such as causal models, that learn abstract variables and causal relationships, can demonstrate increased robustness against changes in the distribution. One reason for this success is the existence and use of Independent Causal Mechanisms (ICMs) representing high-level concepts that only sparsely interact. In this work, we apply two concepts from causality to learn ICMs within LLMs. We develop a new LLM architecture composed of multiple sparsely interacting language modelling modules. We show that such causal constraints can improve out-of-distribution performance on abstract and causal reasoning tasks. We also investigate the level of independence and domain specialisation and show that LLMs rely on pre-trained partially domain-invariant mechanisms resilient to fine-tuning.

Gaël Gendron, Yang Chen, Mitchell Rogers et al.

Better understanding the natural world is a crucial task with a wide range of applications. In environments with close proximity between humans and animals, such as zoos, it is essential to better understand the causes behind animal behaviour and what interventions are responsible for changes in their behaviours. This can help to predict unusual behaviours, mitigate detrimental effects and increase the well-being of animals. There has been work on modelling the dynamics behind swarms of birds and insects but the complex social behaviours of mammalian groups remain less explored. In this work, we propose a method to build behavioural models using causal structure discovery and graph neural networks for time series. We apply this method to a mob of meerkats in a zoo environment and study its ability to predict future actions and model the behaviour distribution at an individual-level and at a group level. We show that our method can match and outperform standard deep learning architectures and generate more realistic data, while using fewer parameters and providing increased interpretability.

Andrey Borro, Patricia J Riddle, Michael W Barley et al.

While systems designed for solving planning tasks vastly outperform Large Language Models (LLMs) in this domain, they usually discard the rich semantic information embedded within task descriptions. In contrast, LLMs possess parametrised knowledge across a wide range of topics, enabling them to leverage the natural language descriptions of planning tasks in their solutions. However, current research in this direction faces challenges in generating correct and executable plans. Furthermore, these approaches depend on the LLM to output solutions in an intermediate language, which must be translated into the representation language of the planning task. We introduce a novel planning method, which leverages the parametrised knowledge of LLMs by using their output as a heuristic for Hill-Climbing Search. This approach is further enhanced by prompting the LLM to generate a solution estimate to guide the search. Our method outperforms the task success rate of similar systems within a common household environment by 22 percentage points, with consistently executable plans. All actions are encoded in their original representation, demonstrating that strong results can be achieved without an intermediate language, thus eliminating the need for a translation step.

Timothy Pistotti, Jason Brown, Michael Witbrock

Recent studies employing Large Language Models (LLMs) to test the Argument from the Poverty of the Stimulus (APS) have yielded contrasting results across syntactic phenomena. This paper investigates the hypothesis that characteristics of the stimuli used in recent studies, including lexical ambiguities and structural complexities, may confound model performance. A methodology is proposed for re-evaluating LLM competence on syntactic prediction, focusing on GPT-2. This involves: 1) establishing a baseline on previously used (both filtered and unfiltered) stimuli, and 2) generating a new, refined dataset using a state-of-the-art (SOTA) generative LLM (Gemini 2.5 Pro Preview) guided by linguistically-informed templates designed to mitigate identified confounds. Our preliminary findings indicate that GPT-2 demonstrates notably improved performance on these refined PG stimuli compared to baselines, suggesting that stimulus quality significantly influences outcomes in surprisal-based evaluations of LLM syntactic competency.

Timothy Pistotti, Jason Brown, Michael Witbrock

Recent studies probing the Argument from the Poverty of the Stimulus (APS) have applied Large Language Models (LLMs) to test the learnability of complex syntax through surprisal-based metrics. However, divergent conclusions raise questions concerning the insights these metrics offer. While Wilcox et al. (2024) used direct minimal pair comparisons (the "wh-effect") to demonstrate that models successfully generalise knowledge of filler-gap dependencies, Lan et al. (2024) used a Difference-in-Differences (DiD) metric and found that models largely fail on parasitic gaps (PGs). This paper argues that the direct minimal pair approach offers greater diagnostic transparency. We demonstrate this by generating a full 8-permutation paradigm of refined PG stimuli and evaluating the GPT-2 model used in previous studies with a systematic Wilcox-style wh-effect analysis. Our results show that GPT-2 succeeds across all four tested conditions, indicating robust knowledge of filler-gap licensing principles even in complex PG environments. This finding, which contrasts with the more ambiguous results from DiD-style metrics, suggests that the choice of evaluation metric is critical for assessing an LLM's syntactic competence.

Yang Chen, Menglin Zou, Jiaqi Zhang et al.

Inverse Reinforcement Learning (IRL) learns a reward function to explain expert demonstrations. Modern IRL methods often use the adversarial (minimax) formulation that alternates between reward and policy optimization, which often lead to unstable training. Recent non-adversarial IRL approaches improve stability by jointly learning reward and policy via energy-based formulations but lack formal guarantees. This work bridges this gap. We first present a unified view showing canonical non-adversarial methods explicitly or implicitly maximize the likelihood of expert behavior, which is equivalent to minimizing the expected return gap. This insight leads to our main contribution: Trust Region Reward Optimization (TRRO), a framework that guarantees monotonic improvement in this likelihood via a Minorization-Maximization process. We instantiate TRRO into Proximal Inverse Reward Optimization (PIRO), a practical and stable IRL algorithm. Theoretically, TRRO provides the IRL counterpart to the stability guarantees of Trust Region Policy Optimization (TRPO) in forward RL. Empirically, PIRO matches or surpasses state-of-the-art baselines in reward recovery, policy imitation with high sample efficiency on MuJoCo and Gym-Robotics benchmarks and a real-world animal behavior modeling task.

Yuchen Su, Yonghua Zhu, Ruofan Wang et al.

Pun generation seeks to creatively modify linguistic elements in text to produce humour or evoke double meanings. It also aims to preserve coherence and contextual appropriateness, making it useful in creative writing and entertainment across various media and contexts. Although pun generation has received considerable attention in computational linguistics, there is currently no dedicated survey that systematically reviews this specific area. To bridge this gap, this paper provides a comprehensive review of pun generation datasets and methods across different stages, including conventional approaches, deep learning techniques, and pre-trained language models. Additionally, we summarise both automated and human evaluation metrics used to assess the quality of pun generation. Finally, we discuss the research challenges and propose promising directions for future work.

Gaël Gendron, Jože M. Rožanec, Michael Witbrock et al.

Causal world models are systems that can answer counterfactual questions about an environment of interest, i.e. predict how it would have evolved if an arbitrary subset of events had been realized differently. It requires understanding the underlying causes behind chains of events and conducting causal inference for arbitrary unseen distributions. So far, this task eludes foundation models, notably large language models (LLMs), which do not have demonstrated causal reasoning capabilities beyond the memorization of existing causal relationships. Furthermore, evaluating counterfactuals in real-world applications is challenging since only the factual world is observed, limiting evaluation to synthetic datasets. We address these problems by explicitly extracting and modeling causal relationships and propose the Causal Cartographer framework. First, we introduce a graph retrieval-augmented generation agent tasked to retrieve causal relationships from data. This approach allows us to construct a large network of real-world causal relationships that can serve as a repository of causal knowledge and build real-world counterfactuals. In addition, we create a counterfactual reasoning agent constrained by causal relationships to perform reliable step-by-step causal inference. We show that our approach can extract causal knowledge and improve the robustness of LLMs for causal reasoning tasks while reducing inference costs and spurious correlations.

Andrey Borro, Patricia J Riddle, Michael W Barley et al.

Recent advancements in the reasoning skills of Large Language Models (LLMs) demonstrate an increase in the ability of LLMs to solve simple planning tasks. However, as long as the driving force behind improved reasoning capability is the size and complexity of the model, the financial and computational costs associated with running them will also increase. This trend raises questions about continued accessibility and whether these improvements will increase at the same pace as models continue to grow in size and expense. We propose two approaches to enhance the reasoning ability of less resource-intensive LLMs. (1) Provide them with a generalised strategy for solving tasks within a given domain, generated by a more resource-intensive LLM. (2) Exploit their cost-effectiveness by iteratively prompting these models to correct errors in their proposed solutions. Our empirical results from planning and mathematical reasoning tasks demonstrate that these methods improve the performance of less resource-intensive LLMs to levels comparable with their more resource-intensive counterparts, at a fraction of the cost. Additionally, we show that the utilisation of generalised strategies in our experiments reduced the cost of the less resource-intensive model by nearly 30 percent on average.

Mitchell Rogers, Kobe Knowles, Gaël Gendron et al.

Deep learning approaches for animal re-identification have had a major impact on conservation, significantly reducing the time required for many downstream tasks, such as well-being monitoring. We propose a method called Recurrence over Video Frames (RoVF), which uses a recurrent head based on the Perceiver architecture to iteratively construct an embedding from a video clip. RoVF is trained using triplet loss based on the co-occurrence of individuals in the video frames, where the individual IDs are unavailable. We tested this method and various models based on the DINOv2 transformer architecture on a dataset of meerkats collected at the Wellington Zoo. Our method achieves a top-1 re-identification accuracy of $49\%$, which is higher than that of the best DINOv2 model ($42\%$). We found that the model can match observations of individuals where humans cannot, and our model (RoVF) performs better than the comparisons with minimal fine-tuning. In future work, we plan to improve these models by using pre-text tasks, apply them to animal behaviour classification, and perform a hyperparameter search to optimise the models further.

Gaël Gendron, Qiming Bao, Michael Witbrock et al.

Large Language Models have shown tremendous performance on a large variety of natural language processing tasks, ranging from text comprehension to common sense reasoning. However, the mechanisms responsible for this success remain opaque, and it is unclear whether LLMs can achieve human-like cognitive capabilities or whether these models are still fundamentally circumscribed. Abstract reasoning is a fundamental task for cognition, consisting of finding and applying a general pattern from few data. Evaluating deep neural architectures on this task could give insight into their potential limitations regarding reasoning and their broad generalisation abilities, yet this is currently an under-explored area. In this paper, we introduce a new benchmark for evaluating language models beyond memorization on abstract reasoning tasks. We perform extensive evaluations of state-of-the-art LLMs, showing that they currently achieve very limited performance in contrast with other natural language tasks, even when applying techniques that have been shown to improve performance on other NLP tasks. We argue that guiding LLM generation to follow causal paths could help improve the generalisation and reasoning abilities of LLMs.

Kobe Knowles, Joshua Bensemann, Diana Benavides-Prado et al.

We introduce a novel architecture, the Neuromodulation Gated Transformer (NGT), which is a simple implementation of neuromodulation in transformers via a multiplicative effect. We compare it to baselines and show that it results in the best average performance on the SuperGLUE benchmark validation sets.

Dave Schneider, Michael Witbrock

In this paper, we discuss Semantic Construction Grammar (SCG), a system developed over the past several years to facilitate translation between natural language and logical representations. Crucially, SCG is designed to support a variety of different methods of representation, ranging from those that are fairly close to the NL structure (e.g. so-called 'logical forms'), to those that are quite different from the NL structure, with higher-order and high-arity relations. Semantic constraints and checks on representations are integral to the process of NL understanding with SCG, and are easily carried out due to the SCG's integration with Cyc's Knowledge Base and inference engine.

Joshua Bensemann, Qiming Bao, Gaël Gendron et al.

Processes occurring in brains, a.k.a. biological neural networks, can and have been modeled within artificial neural network architectures. Due to this, we have conducted a review of research on the phenomenon of blindsight in an attempt to generate ideas for artificial intelligence models. Blindsight can be considered as a diminished form of visual experience. If we assume that artificial networks have no form of visual experience, then deficits caused by blindsight give us insights into the processes occurring within visual experience that we can incorporate into artificial neural networks. This article has been structured into three parts. Section 2 is a review of blindsight research, looking specifically at the errors occurring during this condition compared to normal vision. Section 3 identifies overall patterns from Section 2 to generate insights for computational models of vision. Section 4 demonstrates the utility of examining biological research to inform artificial intelligence research by examining computation models of visual attention relevant to one of the insights generated in Section 3. The research covered in Section 4 shows that incorporating one of our insights into computational vision does benefit those models. Future research will be required to determine whether our other insights are as valuable.

Lin Ni, Qiming Bao, Xiaoxuan Li et al.

Automated question quality rating (AQQR) aims to evaluate question quality through computational means, thereby addressing emerging challenges in online learnersourced question repositories. Existing methods for AQQR rely solely on explicitly-defined criteria such as readability and word count, while not fully utilising the power of state-of-the-art deep-learning techniques. We propose DeepQR, a novel neural-network model for AQQR that is trained using multiple-choice-question (MCQ) datasets collected from PeerWise, a widely-used learnersourcing platform. Along with designing DeepQR, we investigate models based on explicitly-defined features, or semantic features, or both. We also introduce a self-attention mechanism to capture semantic correlations between MCQ components, and a contrastive-learning approach to acquire question representations using quality ratings. Extensive experiments on datasets collected from eight university-level courses illustrate that DeepQR has superior performance over six comparative models.

Ibrahim Abdelaziz, Maxwell Crouse, Bassem Makni et al.

Traditional automated theorem provers have relied on manually tuned heuristics to guide how they perform proof search. Recently, however, there has been a surge of interest in the design of learning mechanisms that can be integrated into theorem provers to improve their performance automatically. In this work, we introduce TRAIL, a deep learning-based approach to theorem proving that characterizes core elements of saturation-based theorem proving within a neural framework. TRAIL leverages (a) an effective graph neural network for representing logical formulas, (b) a novel neural representation of the state of a saturation-based theorem prover in terms of processed clauses and available actions, and (c) a novel representation of the inference selection process as an attention-based action policy. We show through a systematic analysis that these components allow TRAIL to significantly outperform previous reinforcement learning-based theorem provers on two standard benchmark datasets (up to 36% more theorems proved). In addition, to the best of our knowledge, TRAIL is the first reinforcement learning-based approach to exceed the performance of a state-of-the-art traditional theorem prover on a standard theorem proving benchmark (solving up to 17% more problems).

Yang Chen, Libo Zhang, Jiamou Liu et al.

Mean field games (MFGs) provide a mathematically tractable framework for modelling large-scale multi-agent systems by leveraging mean field theory to simplify interactions among agents. It enables applying inverse reinforcement learning (IRL) to predict behaviours of large populations by recovering reward signals from demonstrated behaviours. However, existing IRL methods for MFGs are powerless to reason about uncertainties in demonstrated behaviours of individual agents. This paper proposes a novel framework, Mean-Field Adversarial IRL (MF-AIRL), which is capable of tackling uncertainties in demonstrations. We build MF-AIRL upon maximum entropy IRL and a new equilibrium concept. We evaluate our approach on simulated tasks with imperfect demonstrations. Experimental results demonstrate the superiority of MF-AIRL over existing methods in reward recovery.

Maxwell Crouse, Ibrahim Abdelaziz, Bassem Makni et al.

Automated theorem provers have traditionally relied on manually tuned heuristics to guide how they perform proof search. Deep reinforcement learning has been proposed as a way to obviate the need for such heuristics, however, its deployment in automated theorem proving remains a challenge. In this paper we introduce TRAIL, a system that applies deep reinforcement learning to saturation-based theorem proving. TRAIL leverages (a) a novel neural representation of the state of a theorem prover and (b) a novel characterization of the inference selection process in terms of an attention-based action policy. We show through systematic analysis that these mechanisms allow TRAIL to significantly outperform previous reinforcement-learning-based theorem provers on two benchmark datasets for first-order logic automated theorem proving (proving around 15% more theorems).

Tian Gao, Jie Chen, Vijil Chenthamarakshan et al.

Consider a general machine learning setting where the output is a set of labels or sequences. This output set is unordered and its size varies with the input. Whereas multi-label classification methods seem a natural first resort, they are not readily applicable to set-valued outputs because of the growth rate of the output space; and because conventional sequence generation doesn't reflect sets' order-free nature. In this paper, we propose a unified framework--sequential set generation (SSG)--that can handle output sets of labels and sequences. SSG is a meta-algorithm that leverages any probabilistic learning method for label or sequence prediction, but employs a proper regularization such that a new label or sequence is generated repeatedly until the full set is produced. Though SSG is sequential in nature, it does not penalize the ordering of the appearance of the set elements and can be applied to a variety of set output problems, such as a set of classification labels or sequences. We perform experiments with both benchmark and synthetic data sets and demonstrate SSG's strong performance over baseline methods.

Maxwell Crouse, Achille Fokoue, Maria Chang et al.

Machine learning systems regularly deal with structured data in real-world applications. Unfortunately, such data has been difficult to faithfully represent in a way that most machine learning techniques would expect, i.e. as a real-valued vector of a fixed, pre-specified size. In this work, we introduce a novel approach that compiles structured data into a satisfiability problem which has in its set of solutions at least (and often only) the input data. The satisfiability problem is constructed from constraints which are generated automatically a priori from a given signature, thus trivially allowing for a bag-of-words-esque vector representation of the input to be constructed. The method is demonstrated in two areas, automated reasoning and natural language processing, where it is shown to produce vector representations of natural-language sentences and first-order logic clauses that can be precisely translated back to their original, structured input forms.

Qi Lei, Lingfei Wu, Pin-Yu Chen et al.

Adversarial examples are carefully constructed modifications to an input that completely change the output of a classifier but are imperceptible to humans. Despite these successful attacks for continuous data (such as image and audio samples), generating adversarial examples for discrete structures such as text has proven significantly more challenging. In this paper we formulate the attacks with discrete input on a set function as an optimization task. We prove that this set function is submodular for some popular neural network text classifiers under simplifying assumption. This finding guarantees a $1-1/e$ approximation factor for attacks that use the greedy algorithm. Meanwhile, we show how to use the gradient of the attacked classifier to guide the greedy search. Empirical studies with our proposed optimization scheme show significantly improved attack ability and efficiency, on three different text classification tasks over various baselines. We also use a joint sentence and word paraphrasing technique to maintain the original semantics and syntax of the text. This is validated by a human subject evaluation in subjective metrics on the quality and semantic coherence of our generated adversarial text.

Lingfei Wu, Ian E. H. Yen, Kun Xu et al.

While the celebrated Word2Vec technique yields semantically rich representations for individual words, there has been relatively less success in extending to generate unsupervised sentences or documents embeddings. Recent work has demonstrated that a distance measure between documents called \emph{Word Mover's Distance} (WMD) that aligns semantically similar words, yields unprecedented KNN classification accuracy. However, WMD is expensive to compute, and it is hard to extend its use beyond a KNN classifier. In this paper, we propose the \emph{Word Mover's Embedding } (WME), a novel approach to building an unsupervised document (sentence) embedding from pre-trained word embeddings. In our experiments on 9 benchmark text classification datasets and 22 textual similarity tasks, the proposed technique consistently matches or outperforms state-of-the-art techniques, with significantly higher accuracy on problems of short length.

Ryan Musa, Xiaoyan Wang, Achille Fokoue et al.

Open-domain question answering (QA) is an important problem in AI and NLP that is emerging as a bellwether for progress on the generalizability of AI methods and techniques. Much of the progress in open-domain QA systems has been realized through advances in information retrieval methods and corpus construction. In this paper, we focus on the recently introduced ARC Challenge dataset, which contains 2,590 multiple choice questions authored for grade-school science exams. These questions are selected to be the most challenging for current QA systems, and current state of the art performance is only slightly better than random chance. We present a system that rewrites a given question into queries that are used to retrieve supporting text from a large corpus of science-related text. Our rewriter is able to incorporate background knowledge from ConceptNet and -- in tandem with a generic textual entailment system trained on SciTail that identifies support in the retrieved results -- outperforms several strong baselines on the end-to-end QA task despite only being trained to identify essential terms in the original source question. We use a generalizable decision methodology over the retrieved evidence and answer candidates to select the best answer. By combining query rewriting, background knowledge, and textual entailment our system is able to outperform several strong baselines on the ARC dataset.

Xiaoyan Wang, Pavan Kapanipathi, Ryan Musa et al.

Natural Language Inference (NLI) is fundamental to many Natural Language Processing (NLP) applications including semantic search and question answering. The NLI problem has gained significant attention thanks to the release of large scale, challenging datasets. Present approaches to the problem largely focus on learning-based methods that use only textual information in order to classify whether a given premise entails, contradicts, or is neutral with respect to a given hypothesis. Surprisingly, the use of methods based on structured knowledge -- a central topic in artificial intelligence -- has not received much attention vis-a-vis the NLI problem. While there are many open knowledge bases that contain various types of reasoning information, their use for NLI has not been well explored. To address this, we present a combination of techniques that harness knowledge graphs to improve performance on the NLI problem in the science questions domain. We present the results of applying our techniques on text, graph, and text-to-graph based models, and discuss implications for the use of external knowledge in solving the NLI problem. Our model achieves the new state-of-the-art performance on the NLI problem over the SciTail science questions dataset.

Michael Boratko, Harshit Padigela, Divyendra Mikkilineni et al.

The recent work of Clark et al. introduces the AI2 Reasoning Challenge (ARC) and the associated ARC dataset that partitions open domain, complex science questions into an Easy Set and a Challenge Set. That paper includes an analysis of 100 questions with respect to the types of knowledge and reasoning required to answer them; however, it does not include clear definitions of these types, nor does it offer information about the quality of the labels. We propose a comprehensive set of definitions of knowledge and reasoning types necessary for answering the questions in the ARC dataset. Using ten annotators and a sophisticated annotation interface, we analyze the distribution of labels across the Challenge Set and statistics related to them. Additionally, we demonstrate that although naive information retrieval methods return sentences that are irrelevant to answering the query, sufficient supporting text is often present in the (ARC) corpus. Evaluating with human-selected relevant sentences improves the performance of a neural machine comprehension model by 42 points.

Wei Han, Shiyu Chang, Ding Liu et al.

Advances in image super-resolution (SR) have recently benefited significantly from rapid developments in deep neural networks. Inspired by these recent discoveries, we note that many state-of-the-art deep SR architectures can be reformulated as a single-state recurrent neural network (RNN) with finite unfoldings. In this paper, we explore new structures for SR based on this compact RNN view, leading us to a dual-state design, the Dual-State Recurrent Network (DSRN). Compared to its single state counterparts that operate at a fixed spatial resolution, DSRN exploits both low-resolution (LR) and high-resolution (HR) signals jointly. Recurrent signals are exchanged between these states in both directions (both LR to HR and HR to LR) via delayed feedback. Extensive quantitative and qualitative evaluations on benchmark datasets and on a recent challenge demonstrate that the proposed DSRN performs favorably against state-of-the-art algorithms in terms of both memory consumption and predictive accuracy.