Martin Andrews

Sam Witteveen, Martin Andrews

With the spread of the use of Text2Img diffusion models such as DALL-E 2, Imagen, Mid Journey and Stable Diffusion, one challenge that artists face is selecting the right prompts to achieve the desired artistic output. We present techniques for measuring the effect that specific words and phrases in prompts have, and (in the Appendix) present guidance on the selection of prompts to produce desired effects.

Martin Andrews, Sam Witteveen

Cryptic crossword clues are challenging cognitive tasks, for which new test sets are released on a daily basis by multiple international newspapers. Each cryptic clue contains both the definition of the answer to be placed in the crossword grid (in common with regular crosswords), and `wordplay' that proves that the answer is correct (i.e. a human solver can be confident that an answer is correct without needing crossing words to confirm it). Using an existing cryptic wordplay proving framework (operating on Python proofs created by an LLM), we show that it is possible to distinguish between correct answers and almost-correct ones based upon whether the wordplay `works'.

Martin Andrews

Excellent progress has been made recently in solving ARC Challenge problems. However, it seems that new techniques may be required to push beyond 60% accuracy. Even commercial Large Language Models (LLMs) struggle to 'understand' many of the problems (when given the input and output grids), which makes discovering solutions by LLM-lead program search somewhat futile. In this work, LLM 'understanding' is attempted from a stronger starting position : An LLM is given complete solutions to tasks in code, and then asked to explain how the task is being solved at various levels of abstraction. Specifically, the LLM was given code solutions implemented in arc-dsl-llm (an LLM-legible version of Hodel's arc-dsl to obtain: (a) commented code; (b) code refactored into reusable functional chunks; (c) problem solution steps; and (d) high-level problem-solving tactics. We demonstrate that 'Sparks of Abstraction' can be extracted from the LLM output - in a form that could be used in downstream tasks with Local LLMs eligible to enter the ARC Prize. Both the arc-dsl-llm DSL framework (with the re-engineered solutions) and the Gemini LLM-generated data (along with the generation code) are made Open Source.

Vivek Kalyan, Paul Tan, Shaun Tan et al.

The aim of the CASE 2021 Shared Task 1 (Hürriyetoğlu et al., 2021) was to detect and classify socio-political and crisis event information at document, sentence, cross-sentence, and token levels in a multilingual setting, with each of these subtasks being evaluated separately in each test language. Our submission contained entries in all of the subtasks, and the scores obtained validated our research finding: That the multilingual aspect of the tasks should be embraced, so that modeling and training regimes use the multilingual nature of the tasks to their mutual benefit, rather than trying to tackle the different languages separately. Our code is available at https://github.com/HandshakesByDC/case2021/

Vivek Kalyan, Sam Witteveen, Martin Andrews

Creating explanations for answers to science questions is a challenging task that requires multi-hop inference over a large set of fact sentences. This year, to refocus the Textgraphs Shared Task on the problem of gathering relevant statements (rather than solely finding a single 'correct path'), the WorldTree dataset was augmented with expert ratings of 'relevance' of statements to each overall explanation. Our system, which achieved second place on the Shared Task leaderboard, combines initial statement retrieval; language models trained to predict the relevance scores; and ensembling of a number of the resulting rankings. Our code implementation is made available at https://github.com/mdda/worldtree_corpus/tree/textgraphs_2021

Yew Ken Chia, Sam Witteveen, Martin Andrews

Explainable question answering for science questions is a challenging task that requires multi-hop inference over a large set of fact sentences. To counter the limitations of methods that view each query-document pair in isolation, we propose the LSTM-Interleaved Transformer which incorporates cross-document interactions for improved multi-hop ranking. The LIT architecture can leverage prior ranking positions in the re-ranking setting. Our model is competitive on the current leaderboard for the TextGraphs 2020 shared task, achieving a test-set MAP of 0.5607, and would have gained third place had we submitted before the competition deadline. Our code implementation is made available at https://github.com/mdda/worldtree_corpus/tree/textgraphs_2020

Martin Andrews, Sam Witteveen

Optimizing GPU kernels for high performance is a complex task, often demanding deep architectural knowledge, extensive profiling, and iterative experimentation. This challenge is amplified when targeting newer or less-documented GPU architectures where traditional development aids are scarce. This paper introduces an LLM-powered "GPU Kernel Scientist," an automated methodology for iteratively refining accelerator kernels. Our methodology employs LLMs in a multi-stage, evolutionary process: (a) strategically selecting promising prior code versions as a basis for new iterations; (b) generating hypotheses for optimization experiments, based on existing code and assimilated knowledge from general GPU literature; and (c) autonomously implementing these experiments through code modification and subsequent submission to an external evaluation system, using only observed timing data as performance feedback. We detail how this approach navigates the challenges of the AMD MI300 target architecture and leverages LLMs to compensate for limited domain-specific human expertise. In addition to our results, we present the architectural design, operational workflow, and qualitative insights, highlighting the potential of LLM-driven agents to democratise and accelerate GPU kernel optimization, especially in resource-constrained or rapidly updating hardware environment.

Vivek Kalyan, Martin Andrews

Large Language Model (LLM) agents can leverage multiple turns and tools to solve complex tasks, with prompt-based approaches achieving strong performance. This work demonstrates that Reinforcement Learning (RL) can push capabilities significantly further by learning from experience. Through experiments on a legal document search benchmark, we show that our RL-trained 14 Billion parameter model outperforms frontier class models (85% vs 78% accuracy). In addition, we explore turn-restricted regimes, during training and at test-time, that show these agents achieve better results if allowed to operate over longer multi-turn horizons.

Martin Andrews, Sam Witteveen

Cryptic crossword clues are challenging language tasks for which new test sets are released daily by major newspapers on a global basis. Each cryptic clue contains both the definition of the answer to be placed in the crossword grid (in common with regular crosswords), and 'wordplay' that proves that the answer is correct (i.e. a human solver can be confident that an answer is correct without needing crossing words as confirmation). This work describes an LLM-based reasoning system built from open-licensed components that solves cryptic clues by (i) hypothesising answers; (ii) proposing wordplay explanations; and (iii) using a verifier system that operates on codified reasoning steps. Overall, this system establishes a new state-of-the-art performance on the challenging Cryptonite dataset of clues from The Times and The Telegraph newspapers in the UK. Because each proved solution is expressed in Python, interpretable wordplay reasoning for proven answers is available for inspection.

Sam Witteveen, Martin Andrews

Recently, large language models such as GPT-2 have shown themselves to be extremely adept at text generation and have also been able to achieve high-quality results in many downstream NLP tasks such as text classification, sentiment analysis and question answering with the aid of fine-tuning. We present a useful technique for using a large language model to perform the task of paraphrasing on a variety of texts and subjects. Our approach is demonstrated to be capable of generating paraphrases not only at a sentence level but also for longer spans of text such as paragraphs without needing to break the text into smaller chunks.

Yew Ken Chia, Sam Witteveen, Martin Andrews

The TextGraphs-13 Shared Task on Explanation Regeneration asked participants to develop methods to reconstruct gold explanations for elementary science questions. Red Dragon AI's entries used the language of the questions and explanation text directly, rather than a constructing a separate graph-like representation. Our leaderboard submission placed us 3rd in the competition, but we present here three methods of increasing sophistication, each of which scored successively higher on the test set after the competition close.

Martin Andrews, Sam Witteveen

The recent (2019-02) demonstration of the power of huge language models such as GPT-2 to memorise the answers to factoid questions raises questions about the extent to which knowledge is being embedded directly within these large models. This short paper describes an architecture through which much smaller models can also answer such questions - by making use of 'raw' external knowledge. The contribution of this work is that the methods presented here rely on unsupervised learning techniques, complementing the unsupervised training of the Language Model. The goal of this line of research is to be able to add knowledge explicitly, without extensive training.

Martin Andrews, Yew Ken Chia, Sam Witteveen

Scene graph representations, which form a graph of visual object nodes together with their attributes and relations, have proved useful across a variety of vision and language applications. Recent work in the area has used Natural Language Processing dependency tree methods to automatically build scene graphs. In this work, we present an 'Attention Graph' mechanism that can be trained end-to-end, and produces a scene graph structure that can be lifted directly from the top layer of a standard Transformer model. The scene graphs generated by our model achieve an F-score similarity of 52.21% to ground-truth graphs on the evaluation set using the SPICE metric, surpassing the best previous approaches by 2.5%.

Yew Ken Chia, Sam Witteveen, Martin Andrews

Significant advances have been made in Natural Language Processing (NLP) modelling since the beginning of 2018. The new approaches allow for accurate results, even when there is little labelled data, because these NLP models can benefit from training on both task-agnostic and task-specific unlabelled data. However, these advantages come with significant size and computational costs. This workshop paper outlines how our proposed convolutional student architecture, having been trained by a distillation process from a large-scale model, can achieve 300x inference speedup and 39x reduction in parameter count. In some cases, the student model performance surpasses its teacher on the studied tasks.

Martin Andrews, Sam Witteveen

Relational reasoning is a central component of intelligent behavior, but has proven difficult for neural networks to learn. The Relation Network (RN) module was recently proposed by DeepMind to solve such problems, and demonstrated state-of-the-art results on a number of datasets. However, the RN module scales quadratically in the size of the input, since it calculates relationship factors between every patch in the visual field, including those that do not correspond to entities. In this paper, we describe an architecture that enables relationships to be determined from a stream of entities obtained by an attention mechanism over the input field. The model is trained end-to-end, and demonstrates equivalent performance with greater interpretability while requiring only a fraction of the model parameters of the original RN module.

Martin Andrews

Recent methods for learning vector space representations of words have succeeded in capturing fine-grained semantic and syntactic regularities using vector arithmetic. However, these vector space representations (created through large-scale text analysis) are typically stored verbatim, since their internal structure is opaque. Using word-analogy tests to monitor the level of detail stored in compressed re-representations of the same vector space, the trade-offs between the reduction in memory usage and expressiveness are investigated. A simple scheme is outlined that can reduce the memory footprint of a state-of-the-art embedding by a factor of 10, with only minimal impact on performance. Then, using the same `bit budget', a binary (approximate) factorisation of the same space is also explored, with the aim of creating an equivalent representation with better interpretability.