Research Papers

Concept-based explainable AI is promising as a tool to improve the understanding of complex models at the premises of a given user, viz.\ as a tool for personalized explainability. An important class of concept-based explainability methods is constructed with empirically defined concepts, indirectly defined through a set of positive and negative examples, as in the TCAV approach (Kim et al., 2018). While it is appealing to the user to avoid formal definitions of concepts and their operationalization, it can be challenging to establish relevant concept datasets. Here, we address this challenge using general knowledge graphs (such as, e.g., Wikidata or WordNet) for comprehensive concept definition and present a workflow for user-driven data collection in both text and image domains. The concepts derived from knowledge graphs are defined interactively, providing an opportunity for personalization and ensuring that the concepts reflect the user's intentions. We test the retrieved concept datasets on two concept-based explainability methods, namely concept activation vectors (CAVs) and concept activation regions (CARs) (Crabbe and van der Schaar, 2022). We show that CAVs and CARs based on these empirical concept datasets provide robust and accurate explanations. Importantly, we also find good alignment between the models' representations of concepts and the structure of knowledge graphs, i.e., human representations. This supports our conclusion that knowledge graph-based concepts are relevant for XAI.

In recent years, there has been a notable increase in the deployment of machine learning (ML) models as services (MLaaS) across diverse production software applications. In parallel, explainable AI (XAI) continues to evolve, addressing the necessity for transparency and trustworthiness in ML models. XAI techniques aim to enhance the transparency of ML models by providing insights, in terms of model's explanations, into their decision-making process. Simultaneously, some MLaaS platforms now offer explanations alongside the ML prediction outputs. This setup has elevated concerns regarding vulnerabilities in MLaaS, particularly in relation to privacy leakage attacks such as model extraction attacks (MEA). This is due to the fact that explanations can unveil insights about the inner workings of the model which could be exploited by malicious users. In this work, we focus on investigating how model explanations, particularly counterfactual explanations (CFs), can be exploited for performing MEA within the MLaaS platform. We also delve into assessing the effectiveness of incorporating differential privacy (DP) as a mitigation strategy. To this end, we first propose a novel approach for MEA based on Knowledge Distillation (KD) to enhance the efficiency of extracting a substitute model of a target model exploiting CFs, without any knowledge about the training data distribution by the attacker. Then, we advise an approach for training CF generators incorporating DP to generate private CFs. We conduct thorough experimental evaluations on real-world datasets and demonstrate that our proposed KD-based MEA can yield a high-fidelity substitute model with a reduced number of queries with respect to baseline approaches. Furthermore, our findings reveal that including a privacy layer can allow mitigating the MEA. However, on the account of the quality of CFs, impacts the performance of the explanations.

As artificial intelligence systems become integral across domains, the demand for explainability grows, the called eXplainable artificial intelligence (XAI). Existing efforts primarily focus on generating and evaluating explanations for black-box models while a critical gap in directly enhancing models remains through these evaluations. It is important to consider the potential of this explanation process to improve model quality with a feedback on training as well. XAI may be used to improve model performance while boosting its explainability. Under this view, this paper introduces Transformation - Selective Hidden Input Evaluation for Learning Dynamics (T-SHIELD), a regularization family designed to improve model quality by hiding features of input, forcing the model to generalize without those features. Within this family, we propose the XAI - SHIELD(X-SHIELD), a regularization for explainable artificial intelligence, which uses explanations to select specific features to hide. In contrast to conventional approaches, X-SHIELD regularization seamlessly integrates into the objective function enhancing model explainability while also improving performance. Experimental validation on benchmark datasets underscores X-SHIELD's effectiveness in improving performance and overall explainability. The improvement is validated through experiments comparing models with and without the X-SHIELD regularization, with further analysis exploring the rationale behind its design choices. This establishes X-SHIELD regularization as a promising pathway for developing reliable artificial intelligence regularization.

Using Explainable AI and Hierarchical Planning for Outreach with Robots

Rushang Karia, Jayesh Nagpal, Daksh Dobhal +4

Understanding how robots plan and execute tasks is crucial in today's world, where they are becoming more prevalent in our daily lives. However, teaching non-experts, such as K-12 students, the complexities of robot planning can be challenging. This work presents an open-source platform, JEDAI.Ed, that simplifies the process using a visual interface that abstracts the details of various planning processes that robots use for performing complex mobile manipulation tasks. Using principles developed in the field of explainable AI, this intuitive platform enables students to use a high-level intuitive instruction set to perform complex tasks, visualize them on an in-built simulator, and to obtain helpful hints and natural language explanations for errors. Finally, JEDAI.Ed, includes an adaptive curriculum generation method that provides students with customized learning ramps. This platform's efficacy was tested through a user study with university students who had little to no computer science background. Our results show that JEDAI.Ed is highly effective in increasing student engagement, teaching robotics programming, and decreasing the time need to solve tasks as compared to baselines.

This research presents a novel multimodal data fusion methodology for pain behavior recognition, integrating statistical correlation analysis with human-centered insights. Our approach introduces two key innovations: 1) integrating data-driven statistical relevance weights into the fusion strategy to effectively utilize complementary information from heterogeneous modalities, and 2) incorporating human-centric movement characteristics into multimodal representation learning for detailed modeling of pain behaviors. Validated across various deep learning architectures, our method demonstrates superior performance and broad applicability. We propose a customizable framework that aligns each modality with a suitable classifier based on statistical significance, advancing personalized and effective multimodal fusion. Furthermore, our methodology provides explainable analysis of multimodal data, contributing to interpretable and explainable AI in healthcare. By highlighting the importance of data diversity and modality-specific representations, we enhance traditional fusion techniques and set new standards for recognizing complex pain behaviors. Our findings have significant implications for promoting patient-centered healthcare interventions and supporting explainable clinical decision-making.

Ensuring the security and reliability of machine learning frameworks is crucial for building trustworthy AI-based systems. Fuzzing, a popular technique in secure software development lifecycle (SSDLC), can be used to develop secure and robust software. Popular machine learning frameworks such as PyTorch and TensorFlow are complex and written in multiple programming languages including C/C++ and Python. We propose a dynamic analysis pipeline for Python projects using the Sydr-Fuzz toolset. Our pipeline includes fuzzing, corpus minimization, crash triaging, and coverage collection. Crash triaging and severity estimation are important steps to ensure that the most critical vulnerabilities are addressed promptly. Furthermore, the proposed pipeline is integrated in GitLab CI. To identify the most vulnerable parts of the machine learning frameworks, we analyze their potential attack surfaces and develop fuzz targets for PyTorch, TensorFlow, and related projects such as h5py. Applying our dynamic analysis pipeline to these targets, we were able to discover 3 new bugs and propose fixes for them.

Since their introduction, fuzzy sets and systems have become an important area of research known for its versatility in modelling, knowledge representation and reasoning, and increasingly its potential within the context explainable AI. While the applications of fuzzy systems are diverse, there has been comparatively little advancement in their design from a machine learning perspective. In other words, while representations such as neural networks have benefited from a boom in learning capability driven by an increase in computational performance in combination with advances in their training mechanisms and available tool, in particular gradient descent, the impact on fuzzy system design has been limited. In this paper, we discuss gradient-descent-based optimisation of fuzzy systems, focussing in particular on automatic differentiation -- crucial to neural network learning -- with a view to free fuzzy system designers from intricate derivative computations, allowing for more focus on the functional and explainability aspects of their design. As a starting point, we present a use case in FuzzyR which demonstrates how current fuzzy inference system implementations can be adjusted to leverage powerful features of automatic differentiation tools sets, discussing its potential for the future of fuzzy system design.

Recently, watermarking schemes for large language models (LLMs) have been proposed to distinguish text generated by machines and by humans. The present paper explores philosophical, political, and ethical ramifications of implementing and using watermarking schemes. A definition of authorship that includes both machines (LLMs) and humans is proposed to serve as a backdrop. It is argued that private watermarks may provide private companies with sweeping rights to determine authorship, which is incompatible with traditional standards of authorship determination. Then, possible ramifications of the so-called entropy dependence of watermarking mechanisms are explored. It is argued that entropy may vary for different, socially salient groups. This could lead to group dependent rates at which machine generated text is detected. Specifically, groups more interested in low entropy text may face the challenge that it is harder to detect machine generated text that is of interest to them.

In the current machine translation (MT) landscape, the Transformer architecture stands out as the gold standard, especially for high-resource language pairs. This research delves into its efficacy for low-resource language pairs including both the English$\leftrightarrow$Irish and English$\leftrightarrow$Marathi language pairs. Notably, the study identifies the optimal hyperparameters and subword model type to significantly improve the translation quality of Transformer models for low-resource language pairs. The scarcity of parallel datasets for low-resource languages can hinder MT development. To address this, gaHealth was developed, the first bilingual corpus of health data for the Irish language. Focusing on the health domain, models developed using this in-domain dataset exhibited very significant improvements in BLEU score when compared with models from the LoResMT2021 Shared Task. A subsequent human evaluation using the multidimensional quality metrics error taxonomy showcased the superior performance of the Transformer system in reducing both accuracy and fluency errors compared to an RNN-based counterpart. Furthermore, this thesis introduces adaptNMT and adaptMLLM, two open-source applications streamlined for the development, fine-tuning, and deployment of neural machine translation models. These tools considerably simplify the setup and evaluation process, making MT more accessible to both developers and translators. Notably, adaptNMT, grounded in the OpenNMT ecosystem, promotes eco-friendly natural language processing research by highlighting the environmental footprint of model development. Fine-tuning of MLLMs by adaptMLLM demonstrated advancements in translation performance for two low-resource language pairs: English$\leftrightarrow$Irish and English$\leftrightarrow$Marathi, compared to baselines from the LoResMT2021 Shared Task.

Research interests in the robustness of deep neural networks against domain shifts have been rapidly increasing in recent years. Most existing works, however, focus on improving the accuracy of the model, not the calibration performance which is another important requirement for trustworthy AI systems. Temperature scaling (TS), an accuracy-preserving post-hoc calibration method, has been proven to be effective in in-domain settings, but not in out-of-domain (OOD) due to the difficulty in obtaining a validation set for the unseen domain beforehand. In this paper, we propose consistency-guided temperature scaling (CTS), a new temperature scaling strategy that can significantly enhance the OOD calibration performance by providing mutual supervision among data samples in the source domains. Motivated by our observation that over-confidence stemming from inconsistent sample predictions is the main obstacle to OOD calibration, we propose to guide the scaling process by taking consistencies into account in terms of two different aspects -- style and content -- which are the key components that can well-represent data samples in multi-domain settings. Experimental results demonstrate that our proposed strategy outperforms existing works, achieving superior OOD calibration performance on various datasets. This can be accomplished by employing only the source domains without compromising accuracy, making our scheme directly applicable to various trustworthy AI systems.

With the impressive progress of deep learning, applications relying on machine learning are increasingly being integrated into daily life. However, most deep learning models have an opaque, oracle-like nature making it difficult to interpret and understand their decisions. This problem led to the development of the field known as eXplainable Artificial Intelligence (XAI). One method in this field known as Projective Simulation (PS) models a chain-of-thought as a random walk of a particle on a graph with vertices that have concepts attached to them. While this description has various benefits, including the possibility of quantization, it cannot be naturally used to model thoughts that combine several concepts simultaneously. To overcome this limitation, we introduce Multi-Excitation Projective Simulation (mePS), a generalization that considers a chain-of-thought to be a random walk of several particles on a hypergraph. A definition for a dynamic hypergraph is put forward to describe the agent's training history along with applications to AI and hypergraph visualization. An inductive bias inspired by the remarkably successful few-body interaction models used in quantum many-body physics is formalized for our classical mePS framework and employed to tackle the exponential complexity associated with naive implementations of hypergraphs. We prove that our inductive bias reduces the complexity from exponential to polynomial, with the exponent representing the cutoff on how many particles can interact. We numerically apply our method to two toy environments and a more complex scenario modelling the diagnosis of a broken computer. These environments demonstrate the resource savings provided by an appropriate choice of inductive bias, as well as showcasing aspects of interpretability. A quantum model for mePS is also briefly outlined and some future directions for it are discussed.

This research critically navigates the intricate landscape of AI deception, concentrating on deceptive behaviours of Large Language Models (LLMs). My objective is to elucidate this issue, examine the discourse surrounding it, and subsequently delve into its categorization and ramifications. The essay initiates with an evaluation of the AI Safety Summit 2023 (ASS) and introduction of LLMs, emphasising multidimensional biases that underlie their deceptive behaviours.The literature review covers four types of deception categorised: Strategic deception, Imitation, Sycophancy, and Unfaithful Reasoning, along with the social implications and risks they entail. Lastly, I take an evaluative stance on various aspects related to navigating the persistent challenges of the deceptive AI. This encompasses considerations of international collaborative governance, the reconfigured engagement of individuals with AI, proposal of practical adjustments, and specific elements of digital education.