Frontier research. Production discipline.
Miniml’s leadership includes active academics — University of Edinburgh professors, an Alan Turing Institute Fellow, and an ELLIS Scholar. We work at the boundary of machine learning research and enterprise systems, then carry the methods that hold up under scrutiny into production. Research that ships.
Recent research from the team.
Technical work from Miniml researchers and collaborators — new methods, evaluations, and findings from frontier model development. Each entry links to a short summary on this site, with the full paper one click away.
-
GRADA: Graph-based Reranker against Adversarial Documents Attack
Retrieval Augmented Generation (RAG) frameworks improve the accuracy of large language models (LLMs) by integrating external knowledge from retrieved ...
Read summary -
FLARE: Faithful Logic-Aided Reasoning and Exploration
We introduce Faithful Logic-Aided Reasoning and Exploration (FLARE), a novel interpretable approach for traversing the problem space using task decomp...
Read summary -
DeCoRe: Decoding by Contrasting Retrieval Heads to Mitigate Hallucinations
Large Language Models (LLMs) often hallucinate, producing unfaithful or factually incorrect outputs by misrepresenting the provided context. We propos...
Read summary -
Neurosymbolic Diffusion Models
Neurosymbolic (NeSy) predictors combine neural perception with symbolic reasoning to solve tasks like visual reasoning. However, standard NeSy predict...
Read summary -
MMLongBench: Benchmarking Long-Context Vision-Language Models Effectively and Thoroughly
We introduce MMLongBench, the first benchmark covering a diverse set of long-context vision-language tasks to evaluate long-context vision-language mo...
Read summary -
Neurosymbolic Reasoning Shortcuts under the Independence Assumption
The ubiquitous independence assumption among symbolic concepts in neurosymbolic (NeSy) predictors is a convenient simplification that speeds up probab...
Read summary -
Noiser: Bounded Input Perturbations for Attributing Large Language Models
Feature attribution (FA) methods are common post-hoc approaches that explain how Large Language Models (LLMs) make predictions. Generating faithful at...
Read summary -
An Analysis of Decoding Methods for LLM-based Agents for Faithful Multi-Hop Question Answering
Large Language Models (LLMs) frequently produce factually inaccurate outputs—a phenomenon known as hallucination—which limits their accuracy in knowle...
Read summary -
Q-Filters: Leveraging QK Geometry for Efficient KV Cache Compression
Autoregressive language models rely on a Key-Value (KV) Cache to avoid re-computing past hidden states during generation. As model sizes and context l...
Read summary -
PosterSum: A Multimodal Benchmark for Scientific Poster Summarization
Generating accurate and concise textual summaries from multimodal documents is challenging, especially when dealing with visually complex content like...
Read summary -
Inverse Scaling in Test-Time Compute
We construct evaluation tasks where extending the reasoning length of Large Reasoning Models (LRMs) deteriorates performance, exhibiting an inverse sc...
Read summary -
Lost in Time: Clock and Calendar Understanding Challenges in Multimodal LLMs
Understanding time from visual representations is a fundamental cognitive skill, yet it remains a challenge for multimodal large language models (MLLM...
Read summary -
SynDARin: Synthesising Datasets for Automated Reasoning in Low-Resource Languages
We present SynDARin, a methodology for synthesizing high-quality reasoning datasets in low-resource languages. Our approach combines template-based ge...
Read summary -
Adaptive Computation Modules: Granular Conditional Computation for Efficient Inference
We introduce Adaptive Computation Modules (ACMs) that enable fine-grained conditional computation for more efficient neural network inference. Our app...
Read summary -
Self-Training Large Language Models for Tool-Use Without Demonstrations
We propose a self-training approach that enables large language models to learn tool use without requiring human demonstrations. Our method uses self-...
Read summary -
When Can Proxies Improve the Sample Complexity of Preference Learning?
We provide theoretical and empirical analysis of when proxy rewards can improve sample efficiency in preference learning. Our work establishes conditi...
Read summary -
Is Complex Query Answering Really Complex?
We challenge conventional wisdom about complex query answering by demonstrating that many supposedly complex queries can be solved with surprisingly s...
Read summary -
An Auditing Test to Detect Behavioral Shift in Language Models
We develop a comprehensive auditing framework to detect behavioral shifts in language models across different contexts and time periods. Our approach ...
Read summary -
TuBA: Cross-Lingual Transferability of Backdoor Attacks in LLMs with Instruction Tuning
We investigate the cross-lingual transferability of backdoor attacks in instruction-tuned large language models. Our findings reveal that backdoors ca...
Read summary -
Steering Knowledge Selection Behaviours in LLMs
Read Paper Knowledge Selection in LLMs We investigate how large language models (LLMs) select and utilise knowledge when generating responses. Our analysis reveals that LLMs exhibit systematic biases in knowledge selection, often favouring certain…
Read summary -
Mixtures of In-Context Learners
We introduce Mixtures of In-Context Learners (MiCL), a novel approach that combines multiple in-context learning strategies to improve few-shot perfor...
Read summary -
Low-rank lottery tickets
Low-rank lottery tickets: finding efficient low-rank neural networks via matrix differential equations Read Paper Neural networks deliver exceptional performance but can be impractical for applications with limited hardware or energy resources due to their…
Read summary -
Robust low-rank training via approximate orthonormal constraints
Robust low-rank training via approximate orthonormal constraints Read Paper As models and datasets grow, pruning techniques using low-rank matrix factorizations have become popular for reducing resource demands while maintaining accuracy. However, we find that…
Read summary -
Are We Done with MMLU?
Read Paper Analyzing Flaws in MMLUOur analysis uncovers significant issues with the Massive Multitask Language Understanding (MMLU) benchmark, which is widely used to assess LLMs. We found numerous ground truth errors, with 57% of…
Read summary -
Enhancing AI Model Robustness with Natural Language Explanations
Enhancing AI Model Robustness with Natural Language Explanations Read Paper In this paper, we explore how natural language explanations (NLEs) can improve the robustness of large language models (LLMs) in tasks like natural language…
Read summary -
Probing the Emergence of Cross-lingual Alignment during LLM Training
Probing the Emergence of Cross-lingual Alignment during LLM Training Read Paper Multilingual LLMs excel at zero-shot cross-lingual transfer, likely by aligning languages without parallel sentence supervision. This study uses intrinsic probing to analyze neuron…
Read summary -
Using Natural Language Explanations to Improve Robustness of In-context Learning
Using Natural Language Explanations to Improve Robustness of In-context Learning Read Paper This work explores improving the robustness of LLMs against adversarial inputs by augmenting in-context learning (ICL) with natural language explanations (NLEs). Prompting…
Read summary -
SPARSEFIT: Few-shot Prompting with Sparse Fine-tuning
SPARSEFIT: Few-shot Prompting with Sparse Fine-tuning for Jointly Generating Predictions and Natural Language Explanations Read Paper This work introduces SparseFit, a sparse few-shot fine-tuning strategy for generating natural language explanations (NLEs) with pre-trained language…
Read summary -
Analysing The Impact of Sequence Composition on Language Model Pre-Training
Analysing The Impact of Sequence Composition on Language Model Pre-Training Read Paper Pre-training sequence composition plays a critical role in language model performance. Traditional causal masking can introduce distractions from unrelated documents, hindering effectiveness.…
Read summary -
A Simple and Effective L2 Norm-Based Strategy for KV Cache Compression
A Simple and Effective L2 Norm-Based Strategy for KV Cache Compression Read Paper Deploying large language models (LLMs) is challenging due to the high memory demands of the Key-Value (KV) cache, especially with longer…
Read summary
How the research earns its place.
Most enterprise problems do not need a new method. They need known methods applied with discipline. Research enters our work where it genuinely moves the result — and the bar for that is high.
- Evaluation first. Before a method ships, it is measured against your real data and real failure modes — not a benchmark. The same evaluation rigor we use in research is how we decide what is ready for production.
- Reliability over novelty. A dependable, well-understood approach beats a clever one that drifts. We reach for frontier methods only where a simpler approach genuinely cannot meet the reliability bar.
- Novel methods where they pay. When a problem sits at the edge of what current tools can do, our academic leadership can bring methods from the literature — and the judgment to know when not to.
- Built to own. What we ship is documented, testable, and handed over — so your team can run, extend, and re-evaluate it without us in the room.
Have a problem at the edge of what’s possible?
If your hardest problem needs more than off-the-shelf tools — or you simply want a second opinion grounded in current research — talk to us. A 30-minute conversation with a senior consultant who can tell you what is genuinely solvable today, and what it would take to ship it.
Book a consultation →