Image recognition tasks typically use deep learning and require enormous processing power, thus relying on hardware accelerators like GPUs and FPGAs for fast, timely processing. Failure in real-time image recognition tasks can occur due to incorrect mapping on hardware accelerators, which may lead to timing uncertainty and incorrect behavior. Owing to the increased use of image recognition tasks in safety-critical applications like autonomous driving and medical imaging, it is imperative to assess their robustness to changes in the computational environment as parameters like deep learning frameworks, compiler optimizations for code generation, and hardware devices are not regulated with varying impact on model performance and correctness. In this paper we conduct robustness analysis of four popular image recognition models (MobileNetV2, ResNet101V2, DenseNet121 and InceptionV3) with the ImageNet dataset, assessing the impact of the following parameters in the model's computational environment: (1) deep learning frameworks; (2) compiler optimizations; and (3) hardware devices. We report sensitivity of model performance in terms of output label and inference time for changes in each of these environment parameters. We find that output label predictions for all four models are sensitive to choice of deep learning framework (by up to 57%) and insensitive to other parameters. On the other hand, model inference time was affected by all environment parameters with changes in hardware device having the most effect. The extent of effect was not uniform across models.

Africa has a high student-to-teacher ratio which limits students' access to teachers for learning support such as educational question answering. In this work, we extended Kwame, our previous AI teaching assistant for coding education, adapted it for science education, and deployed it as a web app. Kwame for Science provides passages from well-curated knowledge sources and related past national exam questions as answers to questions from students based on the Integrated Science subject of the West African Senior Secondary Certificate Examination (WASSCE). Furthermore, students can view past national exam questions along with their answers and filter by year, question type (objectives, theory, and practicals), and topics that were automatically categorized by a topic detection model which we developed (91% unweighted average recall). We deployed Kwame for Science in the real world over 8 months and had 750 users across 32 countries (15 in Africa) and 1.5K questions asked. Our evaluation showed an 87.2% top 3 accuracy (n=109 questions) implying that Kwame for Science has a high chance of giving at least one useful answer among the 3 displayed. We categorized the reasons the model incorrectly answered questions to provide insights for future improvements. We also share challenges and lessons with the development, deployment, and human-computer interaction component of such a tool to enable other researchers to deploy similar tools. With a first-of-its-kind tool within the African context, Kwame for Science has the potential to enable the delivery of scalable, cost-effective, and quality remote education to millions of people across Africa.

Due to the extensive growth of information available online, recommender systems play a more significant role in serving people's interests. Traditional recommender systems mostly use an accuracy-focused approach to produce recommendations. Today's research suggests that this single-dimension approach can lead the system to be biased against a series of items with certain attributes. Biased recommendations across groups of items can endanger the interests of item providers along with causing user dissatisfaction with the system. This study aims to manage a new type of intersectional bias regarding the geographical origin and popularity of items in the output of state-of-the-art collaborative filtering recommender algorithms. We introduce an algorithm called MFAIR, a multi-facet post-processing bias mitigation algorithm to alleviate these biases. Extensive experiments on two real-world datasets of movies and books, enriched with the items' continents of production, show that the proposed algorithm strikes a reasonable balance between accuracy and both types of the mentioned biases. According to the results, our proposed approach outperforms a well-known competitor with no or only a slight loss of efficiency.

The approximate nearest neighbor (ANN) search problem is fundamental to efficiently serving many real-world machine learning applications. A number of techniques have been developed for ANN search that are efficient, accurate, and scalable. However, such techniques typically have a number of parameters that affect the speed-recall tradeoff, and exhibit poor performance when such parameters aren't properly set. Tuning these parameters has traditionally been a manual process, demanding in-depth knowledge of the underlying search algorithm. This is becoming an increasingly unrealistic demand as ANN search grows in popularity. To tackle this obstacle to ANN adoption, this work proposes a constrained optimization-based approach to tuning quantization-based ANN algorithms. Our technique takes just a desired search cost or recall as input, and then generates tunings that, empirically, are very close to the speed-recall Pareto frontier and give leading performance on standard benchmarks.

Deploying machine learning models to new tasks is a major challenge despite the large size of the modern training datasets. However, it is conceivable that the training data can be reweighted to be more representative of the new (target) task. We consider the problem of reweighing the training samples to gain insights into the distribution of the target task. Specifically, we formulate a distribution shift model based on the exponential tilt assumption and learn train data importance weights minimizing the KL divergence between labeled train and unlabeled target datasets. The learned train data weights can then be used for downstream tasks such as target performance evaluation, fine-tuning, and model selection. We demonstrate the efficacy of our method on Waterbirds and Breeds benchmarks.

Artificial intelligence caught the world's attention with the rise of ChatGPT. However, this generative AI tool can do more than your homework. It can explain complicated ideas, create social media posts, and so much more! This article will discuss actions you didn't know ChatGPT could do. That way, you could significantly enhance your career and other aspects of daily life.

Swarm robotics provides an approach to the coordination of large numbers of robots inspired by swarm behaviours in nature [1]. The overall behaviours of a swarm are not explicitly engineered in the system. Instead, they are an emergent consequence of the interactions of individual agents with each other and the environment [2]; this poses a challenge to assurance. According to the ISO standard for systems and software engineering vocabulary [3], assurance is defined as "all the planned and systematic activities implemented within the quality system, and demonstrated as needed, to provide adequate confidence that an entity will fulfil requirements for quality". Assurance tasks comprise conformance to standards, verification and validation (V&V), and certification. Assurance criteria for autonomous systems (AS) include both functional and non-functional requirements such as safety [4]. Existing standards and regulations of AS are either implicitly or explicitly based on the V lifecycle model [5], which moves from requirements through design onto implementation and testing before deployment [6, 7]. However, this model is unlikely to be suitable for systems with emergent behaviour (EB); for example through interaction with other agents and the environment, as is the case with swarms. ISO standards have been developed for the service robotics sector (non-industrial) (e.g.

Zero-shot information extraction (IE) aims to build IE systems from the unannotated text. It is challenging due to involving little human intervention. Challenging but worthwhile, zero-shot IE reduces the time and effort that data labeling takes. Recent efforts on large language models (LLMs, e.g., GPT-3, ChatGPT) show promising performance on zero-shot settings, thus inspiring us to explore prompt-based methods. In this work, we ask whether strong IE models can be constructed by directly prompting LLMs. Specifically, we transform the zero-shot IE task into a multi-turn question-answering problem with a two-stage framework (ChatIE). With the power of ChatGPT, we extensively evaluate our framework on three IE tasks: entity-relation triple extract, named entity recognition, and event extraction. Empirical results on six datasets across two languages show that ChatIE achieves impressive performance and even surpasses some full-shot models on several datasets (e.g., NYT11-HRL). We believe that our work could shed light on building IE models with limited resources.

Deep learning models are often trained on distributed, webscale datasets crawled from the internet. In this paper, we introduce two new dataset poisoning attacks that intentionally introduce malicious examples to a model's performance. Our attacks are immediately practical and could, today, poison 10 popular datasets. Our first attack, split-view poisoning, exploits the mutable nature of internet content to ensure a dataset annotator's initial view of the dataset differs from the view downloaded by subsequent clients. By exploiting specific invalid trust assumptions, we show how we could have poisoned 0.01% of the LAION-400M or COYO-700M datasets for just $60 USD. Our second attack, frontrunning poisoning, targets web-scale datasets that periodically snapshot crowd-sourced content -- such as Wikipedia -- where an attacker only needs a time-limited window to inject malicious examples. In light of both attacks, we notify the maintainers of each affected dataset and recommended several low-overhead defenses.

Identifying the main features and learning the causal relationships of a dynamic system from timeseries of sensor data are key problems in many real-world robot applications. In this paper, we propose an extension of a state-of-the-art causal discovery method, PCMCI, embedding an additional feature-selection module based on transfer entropy. Starting from a prefixed set of variables, the new algorithm reconstructs the causal model of the observed system by considering only its main features and neglecting those deemed unnecessary for understanding the evolution of the system. We first validate the method on a toy problem and on synthetic data of brain network, for which the ground-truth models are available, and then on a real-world robotics scenario using a large-scale time-series dataset of human trajectories. The experiments demonstrate that our solution outperforms the previous state-of-the-art technique in terms of accuracy and computational efficiency, allowing better and faster causal discovery of meaningful models from robot sensor data. Keywords: causal discovery, feature selection, time-series, transfer entropy, causal robotics.