Thang Bui

Tuan Dung Nguyen, Yuan-Sen Ting, Ioana Ciucă et al.

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.

Charles O'Neill, Jack Miller, Ioana Ciuca et al.

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.

Charles O'Neill, Yuan-Sen Ting, Ioana Ciuca et al.

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.

Jack Miller, Charles O'Neill, Thang Bui

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.

Charles O'Neill, Tirthankar Ghosal, Roberta Răileanu et al.

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.

Tuan Bui, Oanh Tran, Phuong Nguyen et al.

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.

Charles O'Neill, Thang Bui

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.

Tuan Bui, An Nguyen, Phat Thai et al.

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.

Jack Miller, Patrick Gleeson, Charles O'Neill et al.

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.

Tuan Bui, Trong Le, Phat Thai et al.

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.

Vaden Masrani, Rob Brekelmans, Thang Bui et al.

Many common machine learning methods involve the geometric annealing path, a sequence of intermediate densities between two distributions of interest constructed using the geometric average. While alternatives such as the moment-averaging path have demonstrated performance gains in some settings, their practical applicability remains limited by exponential family endpoint assumptions and a lack of closed form energy function. In this work, we introduce $q$-paths, a family of paths which is derived from a generalized notion of the mean, includes the geometric and arithmetic mixtures as special cases, and admits a simple closed form involving the deformed logarithm function from nonextensive thermodynamics. Following previous analysis of the geometric path, we interpret our $q$-paths as corresponding to a $q$-exponential family of distributions, and provide a variational representation of intermediate densities as minimizing a mixture of $α$-divergences to the endpoints. We show that small deviations away from the geometric path yield empirical gains for Bayesian inference using Sequential Monte Carlo and generative model evaluation using Annealed Importance Sampling.

Rob Brekelmans, Vaden Masrani, Thang Bui et al.

Annealed importance sampling (AIS) is the gold standard for estimating partition functions or marginal likelihoods, corresponding to importance sampling over a path of distributions between a tractable base and an unnormalized target. While AIS yields an unbiased estimator for any path, existing literature has been primarily limited to the geometric mixture or moment-averaged paths associated with the exponential family and KL divergence. We explore AIS using $q$-paths, which include the geometric path as a special case and are related to the homogeneous power mean, deformed exponential family, and $α$-divergence.

Thang Bui, Scott D. Stoller

Attribute-Based Access Control (ABAC) and Relationship-based access control (ReBAC) provide a high level of expressiveness and flexibility that promote security and information sharing, by allowing policies to be expressed in terms of attributes of and chains of relationships between entities. Algorithms for learning ABAC and ReBAC policies from legacy access control information have the potential to significantly reduce the cost of migration to ABAC or ReBAC. This paper presents the first algorithms for mining ABAC and ReBAC policies from access control lists (ACLs) and incomplete information about entities, where the values of some attributes of some entities are unknown. We show that the core of this problem can be viewed as learning a concise three-valued logic formula from a set of labeled feature vectors containing unknowns, and we give the first algorithm (to the best of our knowledge) for that problem.

Thang Bui, Scott D. Stoller

Relationship-based access control (ReBAC) provides a high level of expressiveness and flexibility that promotes security and information sharing, by allowing policies to be expressed in terms of chains of relationships between entities. ReBAC policy mining algorithms have the potential to significantly reduce the cost of migration from legacy access control systems to ReBAC, by partially automating the development of a ReBAC policy. This paper presents new algorithms, called DTRM (Decision Tree ReBAC Miner) and DTRM$^-$, based on decision trees, for mining ReBAC policies from access control lists (ACLs) and information about entities. Compared to state-of-the-art ReBAC mining algorithms, our algorithms are significantly faster, achieve comparable policy quality, and can mine policies in a richer language.

Thang Bui, Scott D. Stoller, Hieu Le

Relationship-based access control (ReBAC) is a flexible and expressive framework that allows policies to be expressed in terms of chains of relationship between entities as well as attributes of entities. ReBAC policy mining algorithms have a potential to significantly reduce the cost of migration from legacy access control systems to ReBAC, by partially automating the development of a ReBAC policy. Existing ReBAC policy mining algorithms support a policy language with a limited set of operators; this limits their applicability. This paper presents a ReBAC policy mining algorithm designed to be both (1) easily extensible (to support additional policy language features) and (2) scalable. The algorithm is based on Bui et al.'s evolutionary algorithm for ReBAC policy mining algorithm. First, we simplify their algorithm, in order to make it easier to extend and provide a methodology that extends it to handle new policy language features. However, extending the policy language increases the search space of candidate policies explored by the evolutionary algorithm, thus causes longer running time and/or worse results. To address the problem, we enhance the algorithm with a feature selection phase. The enhancement utilizes a neural network to identify useful features. We use the result of feature selection to reduce the evolutionary algorithm's search space. The new algorithm is easy to extend and, as shown by our experiments, is more efficient and produces better policies.

Thang Bui, Scott D. Stoller, Jiajie Li

Relationship-based access control (ReBAC) provides a high level of expressiveness and flexibility that promotes security and information sharing. We formulate ReBAC as an object-oriented extension of attribute-based access control (ABAC) in which relationships are expressed using fields that refer to other objects, and path expressions are used to follow chains of relationships between objects. ReBAC policy mining algorithms have potential to significantly reduce the cost of migration from legacy access control systems to ReBAC, by partially automating the development of a ReBAC policy from an existing access control policy and attribute data. This paper presents two algorithms for mining ReBAC policies from access control lists (ACLs) and attribute data represented as an object model: a greedy algorithm guided by heuristics, and a grammar-based evolutionary algorithm. An evaluation of the algorithms on four sample policies and two large case studies demonstrates their effectiveness.

Scott D. Stoller, Thang Bui

Temporal role-based access control (TRBAC) extends role-based access control to limit the times at which roles are enabled. This paper presents a new algorithm for mining high-quality TRBAC policies from timed ACLs (i.e., ACLs with time limits in the entries) and optionally user attribute information. Such algorithms have potential to significantly reduce the cost of migration from timed ACLs to TRBAC. The algorithm is parameterized by the policy quality metric. We consider multiple quality metrics, including number of roles, weighted structural complexity (a generalization of policy size), and (when user attribute information is available) interpretability, i.e., how well role membership can be characterized in terms of user attributes. Ours is the first TRBAC policy mining algorithm that produces hierarchical policies, and the first that optimizes weighted structural complexity or interpretability. In experiments with datasets based on real-world ACL policies, our algorithm is more effective than previous algorithms at optimizing policy quality.

Daniel Hernández-Lobato, José Miguel Hernández-Lobato, Yingzhen Li et al.

A method for large scale Gaussian process classification has been recently proposed based on expectation propagation (EP). Such a method allows Gaussian process classifiers to be trained on very large datasets that were out of the reach of previous deployments of EP and has been shown to be competitive with related techniques based on stochastic variational inference. Nevertheless, the memory resources required scale linearly with the dataset size, unlike in variational methods. This is a severe limitation when the number of instances is very large. Here we show that this problem is avoided when stochastic EP is used to train the model.

José Miguel Hernández-Lobato, Yingzhen Li, Mark Rowland et al.

Black-box alpha (BB-$α$) is a new approximate inference method based on the minimization of $α$-divergences. BB-$α$ scales to large datasets because it can be implemented using stochastic gradient descent. BB-$α$ can be applied to complex probabilistic models with little effort since it only requires as input the likelihood function and its gradients. These gradients can be easily obtained using automatic differentiation. By changing the divergence parameter $α$, the method is able to interpolate between variational Bayes (VB) ($α\rightarrow 0$) and an algorithm similar to expectation propagation (EP) ($α= 1$). Experiments on probit regression and neural network regression and classification problems show that BB-$α$ with non-standard settings of $α$, such as $α= 0.5$, usually produces better predictions than with $α\rightarrow 0$ (VB) or $α= 1$ (EP).