Machine learning models involving discrete latent variables require gradient estimators to facilitate backpropagation in a computationally efficient manner. The most recent addition to the Straight-Through family of estimators, ReinMax, can be viewed from a numerical ODE perspective as incorporating an approximation via Heun's method to reduce bias, but at the cost of high variance. In this work, we introduce the ReinMax-Rao and ReinMax-CV estimators which incorporate Rao-Blackwellisation and control variate techniques into ReinMax to reduce its variance. Our estimators demonstrate superior performance on training variational autoencoders with discrete latent spaces. Furthermore, we investigate the possibility of leveraging alternative numerical methods for constructing more accurate gradient approximations and present an alternative view of ReinMax from a simpler numerical integration perspective.

Multi-hop question answering (QA) requires connecting entities and relations across multiple reasoning steps rather than relying solely on textual understanding. While large language models (LLMs) excel at semantic comprehension, they often produce ungrounded reasoning chains; conversely, purely graph-based approaches lack adaptability across domains. This paper introduces NeuroPath, a hybrid reasoning framework that integrates semantic understanding from LLMs with structured inference over Relational Graph Attention Networks (RGATs). NeuroPath employs a reinforcement learning-based retrieval policy to explore relational paths, coupled with an adaptive stopping mechanism that determines when sufficient evidence has been gathered. This design enables interpretable, verifiable, and efficient multi-hop reasoning. Experimental results on three standard benchmarks, HotpotQA, 2WikiMultihopQA, and MuSiQue, demonstrate that NeuroPath consistently outperforms LLM-only, retrieval-augmented, and hybrid approaches with LLMs and knowledge graph (KG) baselines in both accuracy and reasoning consistency. Our findings highlight the potential of combining neural graph reasoning with language-guided policies to achieve explainable and adaptive multi-hop inference.

Attribute-Based Access Control (ABAC) enables highly expressive and flexible access decisions by considering a wide range of contextual attributes. ABAC policies use logical expressions that combine these attributes, allowing for precise and context-aware control. Algorithms that mine ABAC policies from legacy access control systems can significantly reduce the costs associated with migrating to ABAC. However, a major challenge in this process is handling incomplete entity information, where some attribute values are missing. This paper introduces an approach that enhances the policy mining process by predicting or inferring missing attribute values. This is accomplished by employing a contextual clustering technique that groups entities according to their known attributes, which are then used to analyze and refine authorization decisions. By effectively managing incomplete data, our approach provides security administrators with a valuable tool to improve their attribute data and ensure a smoother, more efficient transition to ABAC.

The conversion of knowledge from natural language (NL) into first-order logic (FOL) plays a crucial role in Knowledge Representation and Cognitive Science, particularly in the context of hybrid Neural-Symbolic models. This process not only facilitates logical knowledge modeling but also enhances the effectiveness of language models and symbolic reasoning systems. The binary nature of FOL—where statements are either true or false—ensures greater consistency and efficiency in these models. However, converting natural language sentences into logical expressions remains a significant challenge due to contextual dependencies, varying levels of complexity, and diverse linguistic constraints. Traditional rule-based or statistical approaches often fail to handle this task effectively. Recently, the emergence of large language models (LLMs) has led to breakthroughs in translation and summarization, yet they continue to face difficulties in this specific conversion task. In this paper, we introduce Text-JEPA (Joint Embedding Predictive Architecture for Text), a self-supervised learning approach trained on textual data. We successfully apply this method to the task of NL-to-FOL conversion. To assess the quality of the conversion process, we refine existing evaluation metrics and propose a comprehensive evaluation framework that incorporates syntactic well-formedness, predicate-level semantic equivalence, and logical equivalence. Our experimental results on benchmark datasets demonstrate that this approach opens a new direction for NL-to-FOL conversion, improving the accuracy of results while maintaining computational efficiency. Furthermore, our model outperformed all tested models, including models of the same size fine-tuned and the large-scale model Gemini 1.5 Flash, highlighting its effectiveness in this domain.

Hyper-relational knowledge graphs (HKGs) are increasingly important in applications where context-rich relationships are essential, such as educational question answering (QA). HKG completion, also known as the link prediction task, aims to infer missing links or facts within these complex graph structures. However, existing HKG completion methods struggle with data sparsity and qualifier complexity challenges that are further amplified in low-resource settings like Vietnamese. In this paper, we propose Attentive-HINGE, a novel extension of the HINGE model that incorporates attention mechanisms to dynamically prioritize relevant qualifiers during inference. Our approach improves hyper-relational reasoning by more effectively addressing both entity and relation prediction tasks. To validate our approach, we first introduce Hyper-EduKG, a new Vietnamese hyper-relational knowledge graph (HKG) dataset curated from university-level educational content, specifically constructed to evaluate our model's effectiveness in a low-resource, domain-specific setting. We then further assess the model's generalizability by evaluating it on two widelyused HKG benchmarks, JF17K and WD50K. Our findings show that Attentive-HINGE not only improves HKG completion but also enables systematic evaluation of knowledge graph quality. This work offers a practical solution for enhancing knowledge graph completion in low-resource languages, paving the way for downstream, closed-domain QA tasks in real-world settings and highlighting the broader utility of attention-based models for hyper-relational reasoning.

Recent advances in large language models (LLMs) have significantly enhanced question-answering (QA) capabilities, particularly in open-domain contexts. However, in closed-domain scenarios such as education, healthcare, and law, users demand not only accurate answers but also transparent reasoning and explainable decision-making processes. While neural-symbolic (NeSy) frameworks have emerged as a promising solution, leveraging LLMs for natural language understanding and symbolic systems for formal reasoning, existing approaches often rely on large-scale models and exhibit inefficiencies in translating natural language into formal logic representations. To address these limitations, we introduce Text-JEPA (Text-based Joint-Embedding Predictive Architecture), a lightweight yet effective framework for converting natural language into first-order logic (NL2FOL). Drawing inspiration from dual-system cognitive theory, Text-JEPA emulates System 1 by efficiently generating logic representations, while the Z3 solver operates as System 2, enabling robust logical inference. To rigorously evaluate the NL2FOL-to-reasoning pipeline, we propose a comprehensive evaluation framework comprising three custom metrics: conversion score, reasoning score, and Spearman rho score, which collectively capture the quality of logical translation and its downstream impact on reasoning accuracy. Empirical results on domain-specific datasets demonstrate that Text-JEPA achieves competitive performance with significantly lower computational overhead compared to larger LLM-based systems. Our findings highlight the potential of structured, interpretable reasoning frameworks for building efficient and explainable QA systems in specialized domains.

Reasoning is essential for closed-domain QA systems in which procedural correctness and policy compliance are critical. While large language models (LLMs) have shown strong performance on many reasoning tasks, recent work reveals that their reasoning traces are often unfaithful - serving more as plausible justifications than as causally grounded derivations. Efforts to combine LLMs with symbolic engines (e.g., Prover9, Z3) have improved reliability but remain limited to static forms of logic, struggling with dynamic, state-based reasoning such as multi-step progressions and conditional transitions. In this paper, we propose MCFR (Model Checking for Formal Reasoning), a neuro-symbolic framework that integrates LLMs with model checking to support property verification. MCFR translates natural language into formal specifications and verifies them over transition models. To support evaluation, we introduce EduMC-QA, a benchmark dataset grounded in real academic procedures. Our results show that MCFR improves reasoning faithfulness and interpretability, offering a viable path toward verifiable QA in high-stakes closed-domain applications. In addition to evaluating MCFR, we compare its performance with state-of-the-art LLMs such as ChatGPT, DeepSeek, and Claude to contextualize its effectiveness.

Inducing-point-based sparse variational Gaussian processes have become the standard workhorse for scaling up GP models. Recent advances show that these methods can be improved by introducing a diagonal scaling matrix to the conditional posterior density given the inducing points. This paper first considers an extension that employs a block-diagonal structure for the scaling matrix, provably tightening the variational lower bound. We then revisit the unifying framework of sparse GPs based on Power Expectation Propagation (PEP) and show that it can leverage and benefit from the new structured approximate posteriors. Through extensive regression experiments, we show that the proposed block-diagonal approximation consistently performs similarly to or better than existing diagonal approximations while maintaining comparable computational costs. Furthermore, the new PEP framework with structured posteriors provides competitive performance across various power hyperparameter settings, offering practitioners flexible alternatives to standard variational approaches.

Generating novel and creative scientific hypotheses is a cornerstone in achieving Artificial General Intelligence. Large language and reasoning models have the potential to aid in the systematic creation, selection, and validation of scientifically informed hypotheses. However, current foundation models often struggle to produce scientific ideas that are both novel and feasible. One reason is the lack of a dedicated dataset that frames Scientific Hypothesis Generation (SHG) as a Natural Language Generation (NLG) task. In this paper, we introduce HypoGen, the first dataset of approximately 5500 structured problem-hypothesis pairs extracted from top-tier computer science conferences structured with a Bit-Flip-Spark schema, where the Bit is the conventional assumption, the Spark is the key insight or conceptual leap, and the Flip is the resulting counterproposal. HypoGen uniquely integrates an explicit Chain-of-Reasoning component that reflects the intellectual process from Bit to Flip. We demonstrate that framing hypothesis generation as conditional language modelling, with the model fine-tuned on Bit-Flip-Spark and the Chain-of-Reasoning (and where, at inference, we only provide the Bit), leads to improvements in the overall quality of the hypotheses. Our evaluation employs automated metrics and LLM judge rankings for overall quality assessment. We show that by fine-tuning on our HypoGen dataset we improve the novelty, feasibility, and overall quality of the generated hypotheses. The HypoGen dataset is publicly available at huggingface.co/datasets/UniverseTBD/hypogen-dr1.

Attribute-Based Access Control (ABAC) provides expressiveness and flexibility, making it a compelling model for enforcing fine-grained access control policies. To facilitate the transition to ABAC, extensive research has been conducted to develop methodologies, frameworks, and tools that assist policy administrators in adapting the model. Despite these efforts, challenges remain in the availability and benchmarking of ABAC datasets. Specifically, there is a lack of clarity on how datasets can be systematically acquired, no standardized benchmarking practices to evaluate existing methodologies and their effectiveness, and limited access to real-world datasets suitable for policy analysis and testing. This paper introduces ABAC Lab, an interactive platform that addresses these challenges by integrating existing ABAC policy datasets with analytical tools for policy evaluation. Additionally, we present two new ABAC datasets derived from real-world case studies. ABAC Lab serves as a valuable resource for both researchers studying ABAC policies and policy administrators seeking to adopt ABAC within their organizations. By offering an environment for dataset exploration and policy analysis, ABAC Lab facilitates research, aids policy administrators in transitioning to ABAC, and promotes a more structured approach to ABAC policy evaluation and development.

Neural networks sometimes exhibit grokking, a phenomenon where perfect or near-perfect performance is achieved on a validation set well after the same performance has been obtained on the corresponding training set. In this workshop paper, we introduce a robust technique for measuring grokking, based on fitting an appropriate functional form. We then use this to investigate the sharpness of transitions in training and validation accuracy under two settings. The first setting is the theoretical framework developed by Levi et al. (2023) where closed form expressions are readily accessible. The second setting is a two-layer MLP trained to predict the parity of bits, with grokking induced by the concealment strategy of Miller et al. (2023). We find that trends between relative grokking gap and grokking sharpness are similar in both settings when using absolute and relative measures of sharpness. Reflecting on this, we make progress toward explaining some trends and identify the need for further study to untangle the various mechanisms which influence the sharpness of grokking.

In today's rapidly evolving landscape of Artificial Intelligence, large language models (LLMs) have emerged as a vibrant research topic. LLMs find applications in various fields and contribute significantly. Despite their powerful language capabilities, similar to pre-trained language models (PLMs), LLMs still face challenges in remembering events, incorporating new information, and addressing domain-specific issues or hallucinations. To overcome these limitations, researchers have proposed Retrieval-Augmented Generation (RAG) techniques, some others have proposed the integration of LLMs with Knowledge Graphs (KGs) to provide factual context, thereby improving performance and delivering more accurate feedback to user queries. Education plays a crucial role in human development and progress. With the technology transformation, traditional education is being replaced by digital or blended education. Therefore, educational data in the digital environment is increasing day by day. Data in higher education institutions are diverse, comprising various sources such as unstructured/structured text, relational databases, web/app-based API access, etc. Constructing a Knowledge Graph from these cross-data sources is not a simple task. This article proposes a method for automatically constructing a Knowledge Graph from multiple data sources and discusses some initial applications (experimental trials) of KG in conjunction with LLMs for question-answering tasks.

In some settings neural networks exhibit a phenomenon known as \textit{grokking}, where they achieve perfect or near-perfect accuracy on the validation set long after the same performance has been achieved on the training set. In this paper, we discover that grokking is not limited to neural networks but occurs in other settings such as Gaussian process (GP) classification, GP regression, linear regression and Bayesian neural networks. We also uncover a mechanism by which to induce grokking on algorithmic datasets via the addition of dimensions containing spurious information. The presence of the phenomenon in non-neural architectures shows that grokking is not restricted to settings considered in current theoretical and empirical studies. Instead, grokking may be possible in any model where solution search is guided by complexity and error.

This paper introduces an efficient and robust method for discovering interpretable circuits in large language models using discrete sparse autoencoders. Our approach addresses key limitations of existing techniques, namely computational complexity and sensitivity to hyperparameters. We propose training sparse autoencoders on carefully designed positive and negative examples, where the model can only correctly predict the next token for the positive examples. We hypothesise that learned representations of attention head outputs will signal when a head is engaged in specific computations. By discretising the learned representations into integer codes and measuring the overlap between codes unique to positive examples for each head, we enable direct identification of attention heads involved in circuits without the need for expensive ablations or architectural modifications. On three well-studied tasks - indirect object identification, greater-than comparisons, and docstring completion - the proposed method achieves higher precision and recall in recovering ground-truth circuits compared to state-of-the-art baselines, while reducing runtime from hours to seconds. Notably, we require only 5-10 text examples for each task to learn robust representations. Our findings highlight the promise of discrete sparse autoencoders for scalable and efficient mechanistic interpretability, offering a new direction for analysing the inner workings of large language models.

Large Language Models (LLMs) hold immense potential to generate synthetic data of high quality and utility, which has numerous applications from downstream model training to practical data utilisation. However, contemporary models, despite their impressive capacities, consistently struggle to produce both coherent and diverse data. To address the coherency issue, we introduce contrastive expert guidance, where the difference between the logit distributions of fine-tuned and base language models is emphasised to ensure domain adherence. In order to ensure diversity, we utilise existing real and synthetic examples as negative prompts to the model. We deem this dual-pronged approach to logit reshaping as STEER: Semantic Text Enhancement via Embedding Repositioning. STEER operates at inference-time and systematically guides the LLMs to strike a balance between adherence to the data distribution (ensuring semantic fidelity) and deviation from prior synthetic examples or existing real datasets (ensuring diversity and authenticity). This delicate balancing act is achieved by dynamically moving towards or away from chosen representations in the latent space. STEER demonstrates improved performance over previous synthetic data generation techniques, exhibiting better balance between data diversity and coherency across three distinct tasks: hypothesis generation, toxic and non-toxic comment generation, and commonsense reasoning task generation. We demonstrate how STEER allows for fine-tuned control over the diversity-coherency trade-off via its hyperparameters, highlighting its versatility.

Large language models excel in many human-language tasks but often falter in highly specialized domains like scholarly astronomy. To bridge this gap, we introduce AstroLLaMA, a 7-billion-parameter model fine-tuned from LLaMA-2 using over 300,000 astronomy abstracts from arXiv. Optimized for traditional causal language modeling, AstroLLaMA achieves a 30% lower perplexity than Llama-2, showing marked domain adaptation. Our model generates more insightful and scientifically relevant text completions and embedding extraction than state-of-the-arts foundation models despite having significantly fewer parameters. AstroLLaMA serves as a robust, domain-specific model with broad fine-tuning potential. Its public release aims to spur astronomy-focused research, including automatic paper summarization and conversational agent development.

In this paper, we tackle the emerging challenge of unintended harmful content generation in Large Language Models (LLMs) with a novel dual-stage optimisation technique using adversarial fine-tuning. Our two-pronged approach employs an adversarial model, fine-tuned to generate potentially harmful prompts, and a judge model, iteratively optimised to discern these prompts. In this adversarial cycle, the two models seek to outperform each other in the prompting phase, generating a dataset of rich examples which are then used for fine-tuning. This iterative application of prompting and fine-tuning allows continuous refinement and improved performance. The performance of our approach is evaluated through classification accuracy on a dataset consisting of problematic prompts not detected by GPT-4, as well as a selection of contentious but unproblematic prompts. We show considerable increase in classification accuracy of the judge model on this challenging dataset as it undergoes the optimisation process. Furthermore, we show that a rudimentary model \texttt{ada} can achieve 13\% higher accuracy on the hold-out test set than GPT-4 after only a few rounds of this process, and that this fine-tuning improves performance in parallel tasks such as toxic comment identification.

In Vietnam nowadays, smart grids have been built and operated recently in some regions when they enable detecting, reacting and pro-acting to changes in usage and multiple issues of the electricity system, and they may have self-healing capabilities. Therefore, study the correctness of the system design of a smart grid has to be carried out carefully before a grid being implement, especially in developing countries like Vietnam.In this research, we proposed a new approach to represent smart grids using Colored Petri Net (CPN). Our approach allows engineers to configure the net dynamically to verify the capacities of the net. The proposed approach also allows engineers to re-configure the system easily to adapt to any change in the grid without re-modelling the system from the grid topology. Additionally, when the state spaces of the nets constructed by the new approach are smaller than that of the conventional modelling approach, the verification for smart grid properties can overcome its ’inherently intractable’ drawback.