Search

Multimodal learning, which involves integrating information from various modalities such as text, images, audio, and video, is pivotal for numerous complex tasks like visual question answering, cross-modal retrieval, and caption generation. Traditional approaches rely on modality-specific encoders and late fusion techniques, which can hinder scalability and flexibility when adapting to new tasks or modalities. To address these limitations, we introduce a novel framework that extends the concept of task reformulation beyond natural language processing (NLP) to multimodal learning. We propose to reformulate diverse multimodal tasks into a unified next-frame prediction problem, allowing a single model to handle different modalities without modality-specific components. This method treats all inputs and outputs as sequential frames in a video, enabling seamless integration of modalities and effective knowledge transfer across tasks. Our approach is evaluated on a range of tasks, including text-to-text, image-to-text, video-to-video, video-to-text, and audio-to-text, demonstrating the model's ability to generalize across modalities with minimal adaptation. We show that task reformulation can significantly simplify multimodal model design across various tasks, laying the groundwork for more generalized multimodal foundation models., Comment: 10 pages, 10 figures

In the fast-paced and volatile financial markets, accurately predicting stock movements based on financial news is critical for investors and analysts. Traditional models often struggle to capture the intricate and dynamic relationships between news events and market reactions, limiting their ability to provide actionable insights. This paper introduces a novel approach leveraging Explainable Artificial Intelligence (XAI) through the development of a Geometric Hypergraph Attention Network (GHAN) to analyze the impact of financial news on market behaviours. Geometric hypergraphs extend traditional graph structures by allowing edges to connect multiple nodes, effectively modelling high-order relationships and interactions among financial entities and news events. This unique capability enables the capture of complex dependencies, such as the simultaneous impact of a single news event on multiple stocks or sectors, which traditional models frequently overlook. By incorporating attention mechanisms within hypergraphs, GHAN enhances the model's ability to focus on the most relevant information, ensuring more accurate predictions and better interpretability. Additionally, we employ BERT-based embeddings to capture the semantic richness of financial news texts, providing a nuanced understanding of the content. Using a comprehensive financial news dataset, our GHAN model addresses key challenges in financial news impact analysis, including the complexity of high-order interactions, the necessity for model interpretability, and the dynamic nature of financial markets. Integrating attention mechanisms and SHAP values within GHAN ensures transparency, highlighting the most influential factors driving market predictions. Empirical validation demonstrates the superior effectiveness of our approach over traditional sentiment analysis and time-series models., Comment: 16 pages, conference

Radiology Report Generation (RRG) has achieved significant progress with the advancements of multimodal generative models. However, the evaluation in the domain suffers from a lack of fair and robust metrics. We reveal that, high performance on RRG with existing lexical-based metrics (e.g. BLEU) might be more of a mirage - a model can get a high BLEU only by learning the template of reports. This has become an urgent problem for RRG due to the highly patternized nature of these reports. In this work, we un-intuitively approach this problem by proposing the Layman's RRG framework, a layman's terms-based dataset, evaluation and training framework that systematically improves RRG with day-to-day language. We first contribute the translated Layman's terms dataset. Building upon the dataset, we then propose a semantics-based evaluation method, which is proved to mitigate the inflated numbers of BLEU and provides fairer evaluation. Last, we show that training on the layman's terms dataset encourages models to focus on the semantics of the reports, as opposed to overfitting to learning the report templates. We reveal a promising scaling law between the number of training examples and semantics gain provided by our dataset, compared to the inverse pattern brought by the original formats. Our code is available at \url{https://github.com/hegehongcha/LaymanRRG}.

With the unprecedented performance being achieved by text-to-image (T2I) diffusion models, T2I customization further empowers users to tailor the diffusion model to new concepts absent in the pre-training dataset, termed subject-driven generation. Moreover, extracting several new concepts from a single image enables the model to learn multiple concepts, and simultaneously decreases the difficulties of training data preparation, urging the disentanglement of multiple concepts to be a new challenge. However, existing models for disentanglement commonly require pre-determined masks or retain background elements. To this end, we propose an attention-guided method, AttenCraft, for multiple concept disentanglement. In particular, our method leverages self-attention and cross-attention maps to create accurate masks for each concept within a single initialization step, omitting any required mask preparation by humans or other models. The created masks are then applied to guide the cross-attention activation of each target concept during training and achieve concept disentanglement. Additionally, we introduce Uniform sampling and Reweighted sampling schemes to alleviate the non-synchronicity of feature acquisition from different concepts, and improve generation quality. Our method outperforms baseline models in terms of image-alignment, and behaves comparably on text-alignment. Finally, we showcase the applicability of AttenCraft to more complicated settings, such as an input image containing three concepts. The project is available at https://github.com/junjie-shentu/AttenCraft.

Next-frame prediction is a useful and powerful method for modelling and understanding the dynamics of video data. Inspired by the empirical success of causal language modelling and next-token prediction in language modelling, we explore the extent to which next-frame prediction serves as a strong foundational learning strategy (analogous to language modelling) for inducing an understanding of the visual world. In order to quantify the specific visual understanding induced by next-frame prediction, we introduce six diagnostic simulation video datasets derived from fundamental physical laws created by varying physical constants such as gravity and mass. We demonstrate that our models trained only on next-frame prediction are capable of predicting the value of these physical constants (e.g. gravity) without having been trained directly to learn these constants via a regression task. We find that the generative training phase alone induces a model state that can predict physical constants significantly better than that of a random model, improving the loss by a factor of between 1.28 to 6.24. We conclude that next-frame prediction shows great promise as a general learning strategy to induce understanding of the many `laws' that govern the visual domain without the need for explicit labelling., Comment: 7 Figures, 12 Pages, 1 Table

Semantic textual similartiy (STS) and information retrieval tasks (IR) tasks have been the two major avenues to record the progress of embedding models in the past few years. Under the emerging Retrieval-augmented Generation (RAG) paradigm, we envision the need to evaluate next-level language understanding abilities of embedding models, and take a conscious look at the reasoning abilities stored in them. Addressing this, we pose the question: Can retrievers solve reasoning problems? By transforming reasoning tasks into retrieval tasks, we find that without specifically trained for reasoning-level language understanding, current state-of-the-art retriever models may still be far from being competent for playing the role of assisting LLMs, especially in reasoning-intensive tasks. Moreover, albeit trained to be aware of instructions, instruction-aware IR models are often better off without instructions in inference time for reasoning tasks, posing an overlooked retriever-LLM behavioral gap for the research community to align. However, recent decoder-based embedding models show great promise in narrowing the gap, highlighting the pathway for embedding models to achieve reasoning-level language understanding. We also show that, although current off-the-shelf re-ranker models fail on these tasks, injecting reasoning abilities into them through fine-tuning still appears easier than doing so to bi-encoders, and we are able to achieve state-of-the-art performance across all tasks by fine-tuning a reranking model. We release Reasoning as Retrieval Benchmark (RAR-b), a holistic suite of tasks and settings to evaluate the reasoning abilities stored in retriever models. RAR-b is available at https://github.com/gowitheflow-1998/RAR-b., Comment: v2, small typo fixes

Achieving an effective fine-grained appearance variation over 2D facial images, whilst preserving facial identity, is a challenging task due to the high complexity and entanglement of common 2D facial feature encoding spaces. Despite these challenges, such fine-grained control, by way of disentanglement is a crucial enabler for data-driven racial bias mitigation strategies across multiple automated facial analysis tasks, as it allows to analyse, characterise and synthesise human facial diversity. In this paper, we propose a novel GAN framework to enable fine-grained control over individual race-related phenotype attributes of the facial images. Our framework factors the latent (feature) space into elements that correspond to race-related facial phenotype representations, thereby separating phenotype aspects (e.g. skin, hair colour, nose, eye, mouth shapes), which are notoriously difficult to annotate robustly in real-world facial data. Concurrently, we also introduce a high quality augmented, diverse 2D face image dataset drawn from CelebA-HQ for GAN training. Unlike prior work, our framework only relies upon 2D imagery and related parameters to achieve state-of-the-art individual control over race-related phenotype attributes with improved photo-realistic output.

Subject-driven text-to-image diffusion models empower users to tailor the model to new concepts absent in the pre-training dataset using a few sample images. However, prevalent subject-driven models primarily rely on single-concept input images, facing challenges in specifying the target concept when dealing with multi-concept input images. To this end, we introduce a textual localized text-to-image model (Texual Localization) to handle multi-concept input images. During fine-tuning, our method incorporates a novel cross-attention guidance to decompose multiple concepts, establishing distinct connections between the visual representation of the target concept and the identifier token in the text prompt. Experimental results reveal that our method outperforms or performs comparably to the baseline models in terms of image fidelity and image-text alignment on multi-concept input images. In comparison to Custom Diffusion, our method with hard guidance achieves CLIP-I scores that are 7.04%, 8.13% higher and CLIP-T scores that are 2.22%, 5.85% higher in single-concept and multi-concept generation, respectively. Notably, our method generates cross-attention maps consistent with the target concept in the generated images, a capability absent in existing models.

Pretrained language models are long known to be subpar in capturing sentence and document-level semantics. Though heavily investigated, transferring perturbation-based methods from unsupervised visual representation learning to NLP remains an unsolved problem. This is largely due to the discreteness of subword units brought by tokenization of language models, limiting small perturbations of inputs to form semantics-preserved positive pairs. In this work, we conceptualize the learning of sentence-level textual semantics as a visual representation learning process. Drawing from cognitive and linguistic sciences, we introduce an unsupervised visual sentence representation learning framework, employing visually-grounded text perturbation methods like typos and word order shuffling, resonating with human cognitive patterns, and enabling perturbation to texts to be perceived as continuous. Our approach is further bolstered by large-scale unsupervised topical alignment training and natural language inference supervision, achieving comparable performance in semantic textual similarity (STS) to existing state-of-the-art NLP methods. Additionally, we unveil our method's inherent zero-shot cross-lingual transferability and a unique leapfrogging pattern across languages during iterative training. To our knowledge, this is the first representation learning method devoid of traditional language models for understanding sentence and document semantics, marking a stride closer to human-like textual comprehension. Our code is available at https://github.com/gowitheflow-1998/Pixel-Linguist

Multi-modal information retrieval (MMIR) is a rapidly evolving field, where significant progress, particularly in image-text pairing, has been made through advanced representation learning and cross-modality alignment research. However, current benchmarks for evaluating MMIR performance in image-text pairing within the scientific domain show a notable gap, where chart and table images described in scholarly language usually do not play a significant role. To bridge this gap, we develop a specialised scientific MMIR (SciMMIR) benchmark by leveraging open-access paper collections to extract data relevant to the scientific domain. This benchmark comprises 530K meticulously curated image-text pairs, extracted from figures and tables with detailed captions in scientific documents. We further annotate the image-text pairs with two-level subset-subcategory hierarchy annotations to facilitate a more comprehensive evaluation of the baselines. We conducted zero-shot and fine-tuning evaluations on prominent multi-modal image-captioning and visual language models, such as CLIP and BLIP. Our analysis offers critical insights for MMIR in the scientific domain, including the impact of pre-training and fine-tuning settings and the influence of the visual and textual encoders. All our data and checkpoints are publicly available at https://github.com/Wusiwei0410/SciMMIR., Comment: camera-ready version for ACL 2024 Findings

In recent years, contrastive learning (CL) has been extensively utilized to recover sentence and document-level encoding capability from pre-trained language models. In this work, we question the length generalizability of CL-based models, i.e., their vulnerability towards length-induced semantic shift. We verify not only that length vulnerability is a significant yet overlooked research gap, but we can devise unsupervised CL methods solely depending on the semantic signal provided by document length. We first derive the theoretical foundations underlying length attacks, showing that elongating a document would intensify the high intra-document similarity that is already brought by CL. Moreover, we found that isotropy promised by CL is highly dependent on the length range of text exposed in training. Inspired by these findings, we introduce a simple yet universal document representation learning framework, LA(SER)$^{3}$: length-agnostic self-reference for semantically robust sentence representation learning, achieving state-of-the-art unsupervised performance on the standard information retrieval benchmark., Comment: Accepted at EMNLP 2023. Our code is publicly available at https://github.com/gowitheflow-1998/LA-SER-cubed

Audio classification plays a crucial role in speech and sound processing tasks with a wide range of applications. There still remains a challenge of striking the right balance between fitting the model to the training data (avoiding overfitting) and enabling it to generalise well to a new domain. Leveraging the transferability of contrastive learning, we introduce Audio Contrastive-based Fine-tuning (AudioConFit), an efficient approach characterised by robust generalisability. Empirical experiments on a variety of audio classification tasks demonstrate the effectiveness and robustness of our approach, which achieves state-of-the-art results in various settings., Comment: Under review

Facial recognition is one of the most academically studied and industrially developed areas within computer vision where we readily find associated applications deployed globally. This widespread adoption has uncovered significant performance variation across subjects of different racial profiles leading to focused research attention on racial bias within face recognition spanning both current causation and future potential solutions. In support, this study provides an extensive taxonomic review of research on racial bias within face recognition exploring every aspect and stage of the face recognition processing pipeline. Firstly, we discuss the problem definition of racial bias, starting with race definition, grouping strategies, and the societal implications of using race or race-related groupings. Secondly, we divide the common face recognition processing pipeline into four stages: image acquisition, face localisation, face representation, face verification and identification, and review the relevant corresponding literature associated with each stage. The overall aim is to provide comprehensive coverage of the racial bias problem with respect to each and every stage of the face recognition processing pipeline whilst also highlighting the potential pitfalls and limitations of contemporary mitigation strategies that need to be considered within future research endeavours or commercial applications alike.

This paper explores deep latent variable models for semi-supervised paraphrase generation, where the missing target pair for unlabelled data is modelled as a latent paraphrase sequence. We present a novel unsupervised model named variational sequence auto-encoding reconstruction (VSAR), which performs latent sequence inference given an observed text. To leverage information from text pairs, we additionally introduce a novel supervised model we call dual directional learning (DDL), which is designed to integrate with our proposed VSAR model. Combining VSAR with DDL (DDL+VSAR) enables us to conduct semi-supervised learning. Still, the combined model suffers from a cold-start problem. To further combat this issue, we propose an improved weight initialisation solution, leading to a novel two-stage training scheme we call knowledge-reinforced-learning (KRL). Our empirical evaluations suggest that the combined model yields competitive performance against the state-of-the-art supervised baselines on complete data. Furthermore, in scenarios where only a fraction of the labelled pairs are available, our combined model consistently outperforms the strong supervised model baseline (DDL) by a significant margin (p <.05; Wilcoxon test). Our code is publicly available at "https://github.com/jialin-yu/latent-sequence-paraphrase".

Incorporating contrastive learning objectives in sentence representation learning (SRL) has yielded significant improvements on many sentence-level NLP tasks. However, it is not well understood why contrastive learning works for learning sentence-level semantics. In this paper, we aim to help guide future designs of sentence representation learning methods by taking a closer look at contrastive SRL through the lens of isotropy, contextualization and learning dynamics. We interpret its successes through the geometry of the representation shifts and show that contrastive learning brings isotropy, and drives high intra-sentence similarity: when in the same sentence, tokens converge to similar positions in the semantic space. We also find that what we formalize as "spurious contextualization" is mitigated for semantically meaningful tokens, while augmented for functional ones. We find that the embedding space is directed towards the origin during training, with more areas now better defined. We ablate these findings by observing the learning dynamics with different training temperatures, batch sizes and pooling methods., Comment: Accepted by ACL 2023 (Findings, long paper)

Reinforcement Learning (RL)-based control system has received considerable attention in recent decades. However, in many real-world problems, such as Batch Process Control, the environment is uncertain, which requires expensive interaction to acquire the state and reward values. In this paper, we present a cost-efficient framework, such that the RL model can evolve for itself in a Virtual Space using the predictive models with only historical data. The proposed framework enables a step-by-step RL model to predict the future state and select optimal actions for long-sight decisions. The main focuses are summarized as: 1) how to balance the long-sight and short-sight rewards with an optimal strategy; 2) how to make the virtual model interacting with real environment to converge to a final learning policy. Under the experimental settings of Fed-Batch Process, our method consistently outperforms the existing state-of-the-art methods.

XAI with natural language processing aims to produce human-readable explanations as evidence for AI decision-making, which addresses explainability and transparency. However, from an HCI perspective, the current approaches only focus on delivering a single explanation, which fails to account for the diversity of human thoughts and experiences in language. This paper thus addresses this gap, by proposing a generative XAI framework, INTERACTION (explaIn aNd predicT thEn queRy with contextuAl CondiTional varIational autO-eNcoder). Our novel framework presents explanation in two steps: (step one) Explanation and Label Prediction; and (step two) Diverse Evidence Generation. We conduct intensive experiments with the Transformer architecture on a benchmark dataset, e-SNLI. Our method achieves competitive or better performance against state-of-the-art baseline models on explanation generation (up to 4.7% gain in BLEU) and prediction (up to 4.4% gain in accuracy) in step one; it can also generate multiple diverse explanations in step two.

Yes - This study investigates the impact of commonplace lossy image compression on face recognition algorithms with regard to the racial characteristics of the subject. We adopt a recently proposed racial phenotype-based bias analysis methodology to measure the effect of varying levels of lossy compression across racial phenotype categories. Additionally, we determine the relationship between chroma-subsampling and race-related phenotypes for recognition performance. Prior work investigates the impact of lossy JPEG compression algorithm on contemporary face recognition performance. However, there is a gap in how this impact varies with different race-related inter-sectional groups and the cause of this impact. Via an extensive experimental setup, we demonstrate that common lossy image compression approaches have a more pronounced negative impact on facial recognition performance for specific racial phenotype categories such as darker skin tones (by up to 34.55\%). Furthermore, removing chroma-subsampling during compression improves the false matching rate (up to 15.95\%) across all phenotype categories affected by the compression, including darker skin tones, wide noses, big lips, and monolid eye categories. In addition, we outline the characteristics that may be attributable as the underlying cause of such phenomenon for lossy compression algorithms such as JPEG.

Surgical workflow anticipation can give predictions on what steps to conduct or what instruments to use next, which is an essential part of the computer-assisted intervention system for surgery, e.g. workflow reasoning in robotic surgery. However, current approaches are limited to their insufficient expressive power for relationships between instruments. Hence, we propose a graph representation learning framework to comprehensively represent instrument motions in the surgical workflow anticipation problem. In our proposed graph representation, we maps the bounding box information of instruments to the graph nodes in the consecutive frames and build inter-frame/inter-instrument graph edges to represent the trajectory and interaction of the instruments over time. This design enhances the ability of our network on modeling both the spatial and temporal patterns of surgical instruments and their interactions. In addition, we design a multi-horizon learning strategy to balance the understanding of various horizons indifferent anticipation tasks, which significantly improves the model performance in anticipation with various horizons. Experiments on the Cholec80 dataset demonstrate the performance of our proposed method can exceed the state-of-the-art method based on richer backbones, especially in instrument anticipation (1.27 v.s. 1.48 for inMAE; 1.48 v.s. 2.68 for eMAE). To the best of our knowledge, we are the first to introduce a spatial-temporal graph representation into surgical workflow anticipation., Comment: Proceedings of the 2022 IEEE-EMBS International Conference on Biomedical and Health Informatics (BHI), 2022

A difficult step in the process of karyotyping is segmenting chromosomes that touch or overlap. In an attempt to automate the process, previous studies turned to Deep Learning methods, with some formulating the task as a semantic segmentation problem. These models treat separate chromosome instances as semantic classes, which we show to be problematic, since it is uncertain which chromosome should be classed as #1 and #2. Assigning class labels based on comparison rules, such as the shorter/longer chromosome alleviates, but does not fully resolve the issue. Instead, we separate the chromosome instances in a second stage, predicting the orientation of the chromosomes by the model and use it as one of the key distinguishing factors of the chromosomes. We demonstrate this method to be effective. Furthermore, we introduce a novel Double-Angle representation that a neural network can use to predict the orientation. The representation maps any direction and its reverse to the same point. Lastly, we present a new expanded synthetic dataset, which is based on Pommier's dataset, but addresses its issues with insufficient separation between its training and testing sets., Comment: Conference for Health, Inference, and Learning (CHIL) 2022 - Invited non-archival presentation

The impressive progress in NLP techniques has been driven by the development of multi-task benchmarks such as GLUE and SuperGLUE. While these benchmarks focus on tasks for one or two input sentences, there has been exciting work in designing efficient techniques for processing much longer inputs. In this paper, we present MuLD: a new long document benchmark consisting of only documents over 10,000 tokens. By modifying existing NLP tasks, we create a diverse benchmark which requires models to successfully model long-term dependencies in the text. We evaluate how existing models perform, and find that our benchmark is much more challenging than their `short document' equivalents. Furthermore, by evaluating both regular and efficient transformers, we show that models with increased context length are better able to solve the tasks presented, suggesting that future improvements in these models are vital for solving similar long document problems. We release the data and code for baselines to encourage further research on efficient NLP models.

Recent work reports disparate performance for intersectional racial groups across face recognition tasks: face verification and identification. However, the definition of those racial groups has a significant impact on the underlying findings of such racial bias analysis. Previous studies define these groups based on either demographic information (e.g. African, Asian etc.) or skin tone (e.g. lighter or darker skins). The use of such sensitive or broad group definitions has disadvantages for bias investigation and subsequent counter-bias solutions design. By contrast, this study introduces an alternative racial bias analysis methodology via facial phenotype attributes for face recognition. We use the set of observable characteristics of an individual face where a race-related facial phenotype is hence specific to the human face and correlated to the racial profile of the subject. We propose categorical test cases to investigate the individual influence of those attributes on bias within face recognition tasks. We compare our phenotype-based grouping methodology with previous grouping strategies and show that phenotype-based groupings uncover hidden bias without reliance upon any potentially protected attributes or ill-defined grouping strategies. Furthermore, we contribute corresponding phenotype attribute category labels for two face recognition tasks: RFW for face verification and VGGFace2 (test set) for face identification., Comment: published in IEEE Winter Conference on Applications of Computer Vision, WACV, 2022

Neural language representation models such as BERT, pre-trained on large-scale unstructured corpora lack explicit grounding to real-world commonsense knowledge and are often unable to remember facts required for reasoning and inference. Natural Language Inference (NLI) is a challenging reasoning task that relies on common human understanding of language and real-world commonsense knowledge. We introduce a new model for NLI called External Knowledge Enhanced BERT (ExBERT), to enrich the contextual representation with real-world commonsense knowledge from external knowledge sources and enhance BERT's language understanding and reasoning capabilities. ExBERT takes full advantage of contextual word representations obtained from BERT and employs them to retrieve relevant external knowledge from knowledge graphs and to encode the retrieved external knowledge. Our model adaptively incorporates the external knowledge context required for reasoning over the inputs. Extensive experiments on the challenging SciTail and SNLI benchmarks demonstrate the effectiveness of ExBERT: in comparison to the previous state-of-the-art, we obtain an accuracy of 95.9% on SciTail and 91.5% on SNLI.

Classifiers tend to propagate biases present in the data on which they are trained. Hence, it is important to understand how the demographic identities of the annotators of comments affect the fairness of the resulting model. In this paper, we focus on the differences in the ways men and women annotate comments for toxicity, investigating how these differences result in models that amplify the opinions of male annotators. We find that the BERT model as-sociates toxic comments containing offensive words with male annotators, causing the model to predict 67.7% of toxic comments as having been annotated by men. We show that this disparity between gender predictions can be mitigated by removing offensive words and highly toxic comments from the training data. We then apply the learned associations between gender and language to toxic language classifiers, finding that models trained exclusively on female-annotated data perform 1.8% better than those trained solely on male-annotated data and that training models on data after removing all offensive words reduces bias in the model by 55.5% while increasing the sensitivity by 0.4%., Comment: Paper is accepted at GeBNLP2021 workshop at ACL-IJCNLP 2021

Deep Learning of neural networks has progressively become more prominent in healthcare with models reaching, or even surpassing, expert accuracy levels. However, these success stories are tainted by concerning reports on the lack of model transparency and bias against some medical conditions or patients' sub-groups. Explainable methods are considered the gateway to alleviate many of these concerns. In this study we demonstrate that the generated explanations are volatile to changes in model training that are perpendicular to the classification task and model structure. This raises further questions about trust in deep learning models for healthcare. Mainly, whether the models capture underlying causal links in the data or just rely on spurious correlations that are made visible via explanation methods. We demonstrate that the output of explainability methods on deep neural networks can vary significantly by changes of hyper-parameters, such as the random seed or how the training set is shuffled. We introduce a measure of explanation consistency which we use to highlight the identified problems on the MIMIC-CXR dataset. We find explanations of identical models but with different training setups have a low consistency: $\approx$ 33% on average. On the contrary, kernel methods are robust against any orthogonal changes, with explanation consistency at 94%. We conclude that current trends in model explanation are not sufficient to mitigate the risks of deploying models in real life healthcare applications., Comment: 9 pages, 5 figures, 3 tables

Explainable machine learning has become increasingly prevalent, especially in healthcare where explainable models are vital for ethical and trusted automated decision making. Work on the susceptibility of deep learning models to adversarial attacks has shown the ease of designing samples to mislead a model into making incorrect predictions. In this work, we propose a model agnostic explainability-based method for the accurate detection of adversarial samples on two datasets with different complexity and properties: Electronic Health Record (EHR) and chest X-ray (CXR) data. On the MIMIC-III and Henan-Renmin EHR datasets, we report a detection accuracy of 77% against the Longitudinal Adversarial Attack. On the MIMIC-CXR dataset, we achieve an accuracy of 88%; significantly improving on the state of the art of adversarial detection in both datasets by over 10% in all settings. We propose an anomaly detection based method using explainability techniques to detect adversarial samples which is able to generalise to different attack methods without a need for retraining., Comment: 13 pages, 6 figures, accepted to ICPR 2020

Disruptive technologies provides unparalleled opportunities to contribute to the identifications of many aspects in pervasive healthcare, from the adoption of the Internet of Things through to Machine Learning (ML) techniques. As a powerful tool, ML has been widely applied in patient-centric healthcare solutions. To further improve the quality of patient care, Electronic Health Records (EHRs) are commonly adopted in healthcare facilities for analysis. It is a crucial task to apply AI and ML to analyse those EHRs for prediction and diagnostics due to their highly unstructured, unbalanced, incomplete, and high-dimensional nature. Dimensionality reduction is a common data preprocessing technique to cope with high-dimensional EHR data, which aims to reduce the number of features of EHR representation while improving the performance of the subsequent data analysis, e.g. classification. In this work, an efficient filter-based feature selection method, namely Curvature-based Feature Selection (CFS), is presented. The proposed CFS applied the concept of Menger Curvature to rank the weights of all features in the given data set. The performance of the proposed CFS has been evaluated in four well-known EHR data sets, including Cervical Cancer Risk Factors (CCRFDS), Breast Cancer Coimbra (BCCDS), Breast Tissue (BTDS), and Diabetic Retinopathy Debrecen (DRDDS). The experimental results show that the proposed CFS achieved state-of-the-art performance on the above data sets against conventional PCA and other most recent approaches. The source code of the proposed approach is publicly available at https://github.com/zhemingzuo/CFS., Comment: Accepted by Technological Forecasting and Social Change; Source code available

Bilinear pooling (BLP) refers to a family of operations recently developed for fusing features from different modalities predominantly developed for VQA models. A bilinear (outer-product) expansion is thought to encourage models to learn interactions between two feature spaces and has experimentally outperformed `simpler' vector operations (concatenation and element-wise-addition/multiplication) on VQA benchmarks. Successive BLP techniques have yielded higher performance with lower computational expense and are often implemented alongside attention mechanisms. However, despite significant progress in VQA, BLP methods have not been widely applied to more recently explored video question answering (video-QA) tasks. In this paper, we begin to bridge this research gap by applying BLP techniques to various video-QA benchmarks, namely: TVQA, TGIF-QA, Ego-VQA and MSVD-QA. We share our results on the TVQA baseline model, and the recently proposed heterogeneous-memory-enchanced multimodal attention (HME) model. Our experiments include both simply replacing feature concatenation in the existing models with BLP, and a modified version of the TVQA baseline to accommodate BLP we name the `dual-stream' model. We find that our relatively simple integration of BLP does not increase, and mostly harms, performance on these video-QA benchmarks. Using recently proposed theoretical multimodal fusion taxonomies, we offer insight into why BLP-driven performance gain for video-QA benchmarks may be more difficult to achieve than in earlier VQA models. We suggest a few additional `best-practices' to consider when applying BLP to video-QA. We stress that video-QA models should carefully consider where the complex representational potential from BLP is actually needed to avoid computational expense on `redundant' fusion., Comment: 16 Pages, 8 Figures, 4 Tables, +Supp Mats

TVQA is a large scale video question answering (video-QA) dataset based on popular TV shows. The questions were specifically designed to require "both vision and language understanding to answer". In this work, we demonstrate an inherent bias in the dataset towards the textual subtitle modality. We infer said bias both directly and indirectly, notably finding that models trained with subtitles learn, on-average, to suppress video feature contribution. Our results demonstrate that models trained on only the visual information can answer ~45% of the questions, while using only the subtitles achieves ~68%. We find that a bilinear pooling based joint representation of modalities damages model performance by 9% implying a reliance on modality specific information. We also show that TVQA fails to benefit from the RUBi modality bias reduction technique popularised in VQA. By simply improving text processing using BERT embeddings with the simple model first proposed for TVQA, we achieve state-of-the-art results (72.13%) compared to the highly complex STAGE model (70.50%). We recommend a multimodal evaluation framework that can highlight biases in models and isolate visual and textual reliant subsets of data. Using this framework we propose subsets of TVQA that respond exclusively to either or both modalities in order to facilitate multimodal modelling as TVQA originally intended., Comment: 10 pages, 4 Figures, 2 Tables, +Supp Mats, BMVC 2020

We consider the task of incorporating real-world commonsense knowledge into deep Natural Language Inference (NLI) models. Existing external knowledge incorporation methods are limited to lexical level knowledge and lack generalization across NLI models, datasets, and commonsense knowledge sources. To address these issues, we propose a novel NLI model-independent neural framework, BiCAM. BiCAM incorporates real-world commonsense knowledge into NLI models. Combined with convolutional feature detectors and bilinear feature fusion, BiCAM provides a conceptually simple mechanism that generalizes well. Quantitative evaluations with two state-of-the-art NLI baselines on SNLI and SciTail datasets in conjunction with ConceptNet and Aristo Tuple KGs show that BiCAM considerably improves the accuracy the incorporated NLI baselines. For example, our BiECAM model, an instance of BiCAM, on the challenging SciTail dataset, improves the accuracy of incorporated baselines by 7.0% with ConceptNet, and 8.0% with Aristo Tuple KG., Comment: Published in Lecture Notes in Computer Science, Springer International Publishing

Despite significant recent progress in the area of Brain-Computer Interface (BCI), there are numerous shortcomings associated with collecting Electroencephalography (EEG) signals in real-world environments. These include, but are not limited to, subject and session data variance, long and arduous calibration processes and predictive generalisation issues across different subjects or sessions. This implies that many downstream applications, including Steady State Visual Evoked Potential (SSVEP) based classification systems, can suffer from a shortage of reliable data. Generating meaningful and realistic synthetic data can therefore be of significant value in circumventing this problem. We explore the use of modern neural-based generative models trained on a limited quantity of EEG data collected from different subjects to generate supplementary synthetic EEG signal vectors, subsequently utilised to train an SSVEP classifier. Extensive experimental analysis demonstrates the efficacy of our generated data, leading to improvements across a variety of evaluations, with the crucial task of cross-subject generalisation improving by over 35% with the use of such synthetic data., Comment: Accepted as a full paper at International Joint Conference on Neural Network (IJCNN) 2019

This paper addresses the challenge of humanoid robot teleoperation in a natural indoor environment via a Brain-Computer Interface (BCI). We leverage deep Convolutional Neural Network (CNN) based image and signal understanding to facilitate both real-time bject detection and dry-Electroencephalography (EEG) based human cortical brain bio-signals decoding. We employ recent advances in dry-EEG technology to stream and collect the cortical waveforms from subjects while they fixate on variable Steady State Visual Evoked Potential (SSVEP) stimuli generated directly from the environment the robot is navigating. To these ends, we propose the use of novel variable BCI stimuli by utilising the real-time video streamed via the on-board robot camera as visual input for SSVEP, where the CNN detected natural scene objects are altered and flickered with differing frequencies (10Hz, 12Hz and 15Hz). These stimuli are not akin to traditional stimuli - as both the dimensions of the flicker regions and their on-screen position changes depending on the scene objects detected. On-screen object selection via such a dry-EEG enabled SSVEP methodology, facilitates the on-line decoding of human cortical brain signals, via a specialised secondary CNN, directly into teleoperation robot commands (approach object, move in a specific direction: right, left or back). This SSVEP decoding model is trained via a priori offline experimental data in which very similar visual input is present for all subjects. The resulting classification demonstrates high performance with mean accuracy of 85% for the real-time robot navigation experiment across multiple test subjects., Comment: Accepted as a full paper at the 2019 International Conference on Robotics and Automation (ICRA)

Dropout is a crucial regularization technique for the Recurrent Neural Network (RNN) models of Natural Language Inference (NLI). However, dropout has not been evaluated for the effectiveness at different layers and dropout rates in NLI models. In this paper, we propose a novel RNN model for NLI and empirically evaluate the effect of applying dropout at different layers in the model. We also investigate the impact of varying dropout rates at these layers. Our empirical evaluation on a large (Stanford Natural Language Inference (SNLI)) and a small (SciTail) dataset suggest that dropout at each feed-forward connection severely affects the model accuracy at increasing dropout rates. We also show that regularizing the embedding layer is efficient for SNLI whereas regularizing the recurrent layer improves the accuracy for SciTail. Our model achieved an accuracy 86.14% on the SNLI dataset and 77.05% on SciTail., Comment: Accepted in International Conference on Artificial Neural Networks, 2018

High Throughput Computing (HTC) provides a convenient mechanism for running thousands of tasks. Many HTC systems exploit computers which are provisioned for other purposes by utilising their idle time - volunteer computing. This has great advantages as it gives access to vast quantities of computational power for little or no cost. The downside is that running tasks are sacrificed if the computer is needed for its primary use. Normally terminating the task which must be restarted on a different computer - leading to wasted energy and an increase in task completion time. We demonstrate, through the use of simulation, how we can reduce this wasted energy by targeting tasks at computers less likely to be needed for primary use, predicting this idle time through machine learning. By combining two machine learning approaches, namely Random Forest and MultiLayer Perceptron, we save 51.4% of the energy without significantly affecting the time to complete tasks., Comment: Accepted for publication at THE 9th international Green and sustainable computing Conference, Technically Co-sponsored by IEEE Computer Society & STC Sustainable Computing, October 22-24, Pittsburgh, PA, USA

In this paper, we propose a novel Convolutional Neural Network (CNN) approach for the classification of raw dry-EEG signals without any data pre-processing. To illustrate the effectiveness of our approach, we utilise the Steady State Visual Evoked Potential (SSVEP) paradigm as our use case. SSVEP can be utilised to allow people with severe physical disabilities such as Complete Locked-In Syndrome or Amyotrophic Lateral Sclerosis to be aided via BCI applications, as it requires only the subject to fixate upon the sensory stimuli of interest. Here we utilise SSVEP flicker frequencies between 10 to 30 Hz, which we record as subject cortical waveforms via the dry-EEG headset. Our proposed end-to-end CNN allows us to automatically and accurately classify SSVEP stimulation directly from the dry-EEG waveforms. Our CNN architecture utilises a common SSVEP Convolutional Unit (SCU), comprising of a 1D convolutional layer, batch normalization and max pooling. Furthermore, we compare several deep learning neural network variants with our primary CNN architecture, in addition to traditional machine learning classification approaches. Experimental evaluation shows our CNN architecture to be significantly better than competing approaches, achieving a classification accuracy of 96% whilst demonstrating superior cross-subject performance and even being able to generalise well to unseen subjects whose data is entirely absent from the training process., Comment: Accepted as a full paper at the 2018 IEEE International Conference on Systems, Man, and Cybernetics (SMC2018)

In This paper we present a novel approach to spam filtering and demonstrate its applicability with respect to SMS messages. Our approach requires minimum features engineering and a small set of la- belled data samples. Features are extracted using topic modelling based on latent Dirichlet allocation, and then a comprehensive data model is created using a Stacked Denoising Autoencoder (SDA). Topic modelling summarises the data providing ease of use and high interpretability by visualising the topics using word clouds. Given that the SMS messages can be regarded as either spam (unwanted) or ham (wanted), the SDA is able to model the messages and accurately discriminate between the two classes without the need for a pre-labelled training set. The results are compared against the state-of-the-art spam detection algorithms with our proposed approach achieving over 97% accuracy which compares favourably to the best reported algorithms presented in the literature., Comment: Paper was accepted to the 25th International Conference on Artificial Neural Networks (ICANN 2016)