General-purpose Tasks Research Articles

A benchmark and a multi-stage pipeline for classifying underwater videos at scale

ABSTRACT Standardised benchmarks have been instrumental in driving the recent progress in computer vision. However, most benchmarks are designed for general-purpose tasks, covering multiple different topics and classes but are limited to the needs of specialised tasks. For example, when performing 3D reconstruction of corals, researchers need to record footage of coral with multiple camera angles. Due to the limited availability of such videos in standard datasets, the ability to reconstruct 3D coral models from public videos would alleviate this problem since it would allow researchers to tap into the vast scope of online content. Thus, one could use machine learning to sift through the immense amounts of content and automatically identify suitable videos for 3D reconstruction. In this work, we introduce a new benchmark that uses amateur footage queried from the YouTube-8 M dataset where each video has been manually labelled for undersea, coral, and multiple camera angles. Furthermore, we construct a three-stage pipeline of machine learning models with the purpose of identifying suitable videos for the 3D reconstruction of coral from the public domain. We instantiate the pipeline with state-of-the-art video classification methods and evaluate their performance on the benchmark, identifying their shortcomings and avenues for future research.

A data and knowledge driven autonomous intelligent manufacturing system for intelligent factories

The non-ferrous metal industry is encountering several challenges, including production efficiency, manufacturing information fragmentation, and human health problems, which highlights the importance of implementing autonomous intelligent manufacturing systems (AIMS). Recently, the foundation model like GPT-4, has garnered attentions due to its exceptional capabilities and proficiency in diverse domains and tasks, facilitating the realization of AIMS. However, the existing foundation models can only address basic general-purpose tasks and are difficult to use for industrial applications. In this paper, we propose a data and knowledge driven AIMS with industrial-generative pretrained Transformer (Industrial-GPT) for intelligent factories. The paradigms and architecture of autonomous intelligent factories are firstly defined. Then, we explore the mechanism with knowledge graph, digital twin, and Industrial-GPT, including multi-level autonomous perception, cross layer and domain cognition, and event-driven collaborative decision-making. Finally, the detailed case study is based on the cooperation with a zinc smelting intelligent factory to achieve networked collaborative manufacturing, and explores the theory and realization mechanism of AIMS on a small scale. We explore the experimental analyses, evaluation mechanisms and platform applications of AIMS at the workshop level. We believe this will help to realize larger scale AIMS in the future.

General Purpose Artificial Intelligence Systems (GPAIS): Properties, definition, taxonomy, societal implications and responsible governance

Most applications of Artificial Intelligence (AI) are designed for a confined and specific task. However, there are many scenarios that call for a more general AI, capable of solving a wide array of tasks without being specifically designed for them. The term General Purpose Artificial Intelligence Systems (GPAIS) has been defined to refer to these AI systems. To date, the possibility of an Artificial General Intelligence, powerful enough to perform any intellectual task as if it were human, or even improve it, has remained an aspiration, fiction, and considered a risk for our society. Whilst we might still be far from achieving that, GPAIS is a reality and sitting at the forefront of AI research.This work discusses existing definitions for GPAIS and proposes a new definition that allows for a gradual differentiation among types of GPAIS according to their properties and limitations. We distinguish between closed-world and open-world GPAIS, characterising their degree of autonomy and ability based on several factors such as adaptation to new tasks, competence in domains not intentionally trained for, ability to learn from few data, or proactive acknowledgement of their own limitations. We then propose a taxonomy of approaches to realise GPAIS, describing research trends such as the use of AI techniques to improve another AI (commonly referred to as AI-powered AI) or (single) foundation models. As a prime example, we delve into generative AI (GenAI), aligning them with the terms and concepts presented in the taxonomy. Similarly, we explore the challenges and prospects of multi-modality, which involves fusing various types of data sources to expand the capabilities of GPAIS. Through the proposed definition and taxonomy, our aim is to facilitate research collaboration across different areas that are tackling general purpose tasks, as they share many common aspects. Finally, with the goal of providing a holistic view of GPAIS, we discuss the current state of GPAIS, its prospects, implications for our society, and the need for regulation and governance of GPAIS to ensure their responsible and trustworthy development.

Deep learning-enabled natural language processing to identify directional pharmacokinetic drug-drug interactions.

During drug development, it is essential to gather information about the change of clinical exposure of a drug (object) due to the pharmacokinetic (PK) drug-drug interactions (DDIs) with another drug (precipitant). While many natural language processing (NLP) methods for DDI have been published, most were designed to evaluate if (and what kind of) DDI relationships exist in the text, without identifying the direction of DDI (object vs. precipitant drug). Here we present a method for the automatic identification of the directionality of a PK DDI from literature or drug labels. We reannotated the Text Analysis Conference (TAC) DDI track 2019 corpus for identifying the direction of a PK DDI and evaluated the performance of a fine-tuned BioBERT model on this task by following the training and validation steps prespecified by TAC. This initial attempt showed the model achieved an F-score of 0.82 in identifying sentences as containing PK DDI and an F-score of 0.97 in identifying object versus precipitant drugs in those sentences. Despite a growing list of NLP methods for DDI extraction, most of them use a common set of corpora to perform general purpose tasks (e.g., classifying a sentence into one of several fixed DDI categories). There is a lack of coordination between the drug development and biomedical informatics method development community to develop corpora and methods to perform specific tasks (e.g., extract clinical exposure changes due to PK DDI). We hope that our effort can encourage such a coordination so that more "fit for purpose" NLP methods could be developed and used to facilitate the drug development process.

Cost-Effective Full-Color 3D Dental Imaging Based on Close-Range Photogrammetry.

Dental imaging plays a crucial role in clinical dental practice. Conventional 2D dental imaging serves general-purpose tasks, such as patient documentation, while high-precision 3D dental scanning is tailored for specialized procedures, such as orthodontics and implant surgeries. In this study, we aimed to develop a cost-effective 3D imaging technique that could bridge the gap between conventional dental photography and high-precision 3D dental scanning, with the goal of improving patient dental care. We developed a 3D imaging technique based on close-range photogrammetry and termed it close-range photogrammetry-based dental imaging (CPDI). We evaluated this technique on both in vitro dental models and in vivo teeth. For dental models, we conducted a parametric study to examine the effects of the depth of field and specular reflection on reconstruction quality. We showed that the optimal results were achieved with an f/5.6 lens and without a circular polarizer for reflection suppression. This configuration generated 3D scans with 57.7 ± 3.2% and 82.4 ± 2.7% of reconstructed points falling within ±0.1 mm and ±0.2 mm error margins, respectively. With such accuracy, these 3D dental models can faithfully represent dental morphology and features. During in vivo imaging, we were able to reconstruct high-quality 3D models of the anterior arch, further demonstrating its clinical relevance. The reconstructed models carry both 3D shapes and detail full-color surface textures, which positions CPDI as a versatile imaging tool in different areas of clinical dental care.

Assessment of Resident and AI Chatbot Performance on the University of Toronto Family Medicine Residency Progress Test: Comparative Study.

Large language model (LLM)-based chatbots are evolving at an unprecedented pace with the release of ChatGPT, specifically GPT-3.5, and its successor, GPT-4. Their capabilities in general-purpose tasks and language generation have advanced to the point of performing excellently on various educational examination benchmarks, including medical knowledge tests. Comparing the performance of these 2 LLM models to that of Family Medicine residents on a multiple-choice medical knowledge test can provide insights into their potential as medical education tools. This study aimed to quantitatively and qualitatively compare the performance of GPT-3.5, GPT-4, and Family Medicine residents in a multiple-choice medical knowledge test appropriate for the level of a Family Medicine resident. An official University of Toronto Department of Family and Community Medicine Progress Test consisting of multiple-choice questions was inputted into GPT-3.5 and GPT-4. The artificial intelligence chatbot's responses were manually reviewed to determine the selected answer, response length, response time, provision of a rationale for the outputted response, and the root cause of all incorrect responses (classified into arithmetic, logical, and information errors). The performance of the artificial intelligence chatbots were compared against a cohort of Family Medicine residents who concurrently attempted the test. GPT-4 performed significantly better compared to GPT-3.5 (difference 25.0%, 95% CI 16.3%-32.8%; McNemar test: P<.001); it correctly answered 89/108 (82.4%) questions, while GPT-3.5 answered 62/108 (57.4%) questions correctly. Further, GPT-4 scored higher across all 11 categories of Family Medicine knowledge. In 86.1% (n=93) of the responses, GPT-4 provided a rationale for why other multiple-choice options were not chosen compared to the 16.7% (n=18) achieved by GPT-3.5. Qualitatively, for both GPT-3.5 and GPT-4 responses, logical errors were the most common, while arithmetic errors were the least common. The average performance of Family Medicine residents was 56.9% (95% CI 56.2%-57.6%). The performance of GPT-3.5 was similar to that of the average Family Medicine resident (P=.16), while the performance of GPT-4 exceeded that of the top-performing Family Medicine resident (P<.001). GPT-4 significantly outperforms both GPT-3.5 and Family Medicine residents on a multiple-choice medical knowledge test designed for Family Medicine residents. GPT-4 provides a logical rationale for its response choice, ruling out other answer choices efficiently and with concise justification. Its high degree of accuracy and advanced reasoning capabilities facilitate its potential applications in medical education, including the creation of exam questions and scenarios as well as serving as a resource for medical knowledge or information on community services.

Fractal Parallel Computing

As machine learning (ML) becomes the prominent technology for many emerging problems, dedicated ML computers are being developed at a variety of scales, from clouds to edge devices. However, the heterogeneous, parallel, and multilayer characteristics of conventional ML computers concentrate the cost of development on the software stack, namely, ML frameworks, compute libraries, and compilers, which limits the productivity of new ML computers. Fractal von Neumann architecture (FvNA) is proposed to address the programming productivity issue for ML computers. FvNA is scale-invariant to program, thus making the development of a family of scaled ML computers as easy as a single node. In this study, we generalize FvNA to the field of general-purpose parallel computing. We model FvNA as an abstract parallel computer, referred to as the fractal parallel machine (FPM), to demonstrate several representative general-purpose tasks that are efficiently programmable. FPM limits the entropy of programming by applying constraints on the control pattern of the parallel computing systems. However, FPM is still general-purpose and cost-optimal. We settle some preliminary results showing that FPM is as powerful as many fundamental parallel computing models such as BSP and alternating Turing machine. Therefore, FvNA is also generally applicable to various fields other than ML.

Deep learning for scanning electron microscopy: Synthetic data for the nanoparticles detection

Deep learning algorithms are one of most rapid developing fields into the modern computation technologies. One of the bottlenecks into the implementation of such advaced algorithms is their requirement for a large amount of manually-labelled data for training. For the general-purpose tasks, such as general purpose image classification/detection the huge images datasets are already labelled and collected. For more subject specific tasks (such as electron microscopy images treatment), no labelled data available. Here I demonstrate that a deep learning network can be successfully trained for nanoparticles detection using semi-synthetic data. The real SEM images were used as a textures for rendered nanoparticles at the surface. Training of RetinaNet architecture using transfer learning can be helpful for the large-scale particle distribution analysis. Beyond such applications, the presented approach might be applicable to other tasks, such as image segmentation.

Assessing the quality of answers autonomously in community question–answering

Community question–answering (CQA) has become a popular method of online information seeking. Within these services, peers ask questions and create answers to those questions. For some time, content repositories created through CQA sites have widely supported general-purpose tasks; however, they can also be used as online digital libraries that satisfy specific needs related to education. Horizontal CQA services, such as Yahoo! Answers, and vertical CQA services, such as Brainly, aim to help students improve their learning process via Q&A exchanges. In addition, Stack Overflow—another vertical CQA—serves a similar purpose but specifically focuses on topics relevant to programmers. Receiving high-quality answer(s) to a posed CQA query is a critical factor to both user satisfaction and supported learning in these services. This process can be impeded when experts do not answer questions and/or askers do not have the knowledge and skills needed to evaluate the quality of the answers they receive. Such circumstances may cause learners to construct a faulty knowledge base by applying inaccurate information acquired from online sources. Though site moderators could alleviate this problem by surveying answer quality, their subjective assessments may cause evaluations to be inconsistent. Another potential solution lies in human assessors, though they may also be insufficient due to the large amount of content available on a CQA site. The following study addresses these issues by proposing a framework for automatically assessing answer quality. We accomplish this by integrating different groups of features—personal, community-based, textual, and contextual—to build a classification model and determine what constitutes answer quality. We collected more than 10 million educational answers posted by more than 3 million users on Brainly and 7.7 million answers on Stack Overflow to test this evaluation framework. The experiments conducted on these data sets show that the model using random forest achieves high accuracy in identifying high-quality answers. Findings also indicate that personal and community-based features have more prediction power in assessing answer quality. Additionally, other key metrics such as F1-score and area under ROC curve achieve high values with our approach. The work reported here can be useful in many other contexts that strive to provide automatic quality assessment in a digital repository.

Testing Subjective Preference and Map Use Performance: Use of Web Maps for Decision Making in the Public Health Sector

Maps support many planning activities, from general purpose tasks to in-depth spatial analyses. However, preferences and intuitions, which occur in all human interactions, are often disregarded in use and user research in cartography, and there is evidences that analysis of these factors is desirable. It is commonly believed that there is a positive correlation between subjective preference and objective performance, and there is agreement about the main role of media in map comprehension. To investigate this topic, quantitative experiments were performed, based on a real situation in which students need to make decisions about public health care management in a city. Users were asked about their preferences with respect to map type, and performances in carrying out tasks were measured. The results indicate that for simple visual comparison tasks, the proposed Web map framework was adequate. However, this was not the case for reasoning tasks, where weak performances were registered. Also, user preference among visual variables seemed to be unrelated to better performance, and since performances were poor, the important role played by interface usability and attractiveness in map use is verified.

Evaluation of direct manipulation using finger tracking for complex tasks in an immersive cube

A solution for interaction using finger tracking in a cubic immersive virtual reality system (or immersive cube) is presented. Rather than using a traditional wand device, users can manipulate objects with fingers of both hands in a close-to-natural manner for moderately complex, general purpose tasks. Our solution couples finger tracking with a real-time physics engine, combined with a heuristic approach for hand manipulation, which is robust to tracker noise and simulation instabilities. A first study has been performed to evaluate our interface, with tasks involving complex manipulations, such as balancing objects while walking in the cube. The user's finger-tracked manipulation was compared to manipulation with a 6 degree-of-freedom wand (or flystick), as well as with carrying out the same task in the real world. Users were also asked to perform a free task, allowing us to observe their perceived level of presence in the scene. Our results show that our approach provides a feasible interface for immersive cube environments and is perceived by users as being closer to the real experience compared to the wand. However, the wand outperforms direct manipulation in terms of speed and precision. We conclude with a discussion of the results and implications for further research.

Cache isolation for virtualization of mixed general-purpose and real-time systems

Shared cache has been widely used to lessen the communication overhead between multi-core processors and off-chip main memories. In a virtualization execution environment, real-time tasks supported by real-time operating systems (RTOS) need to share the last level cache with other general purpose tasks (supported by general purpose operating systems (GPOS) or execute on bare metal). Since current hypervisors are not aware of the attributes of tasks in the virtual machines, real-time tasks cannot be adequately supported by the hypervisor. Therefore, real-time task will suffer from many cache misses, and the response of a virtual machine may be poor due to cache pollution, especially if real-time tasks compete for cache resources with heavy general-purpose tasks. To address this issue, we propose two cache optimization management policies. One is cache partitioning (CAP) - which isolates the cache statically for real-time tasks and general-purpose tasks using page coloring. Hence the predictability of cache usage is greatly increased. CAP reduces both cache pollution, and the amount of cache space that real-time tasks use, even if other cache is not used by general purpose tasks. We also study allocating different proportions of the cache for real-time tasks and find that, a proper cache ratio is optimal to guarantee response time. Due to this, another dynamic cache optimization policy is proposed, called page table prefetching (PTP). PTP employs a daemon-like thread to maintain the cache, so as to make the contents used by real-time tasks be in cache. PTP is implemented by utilizing a dedicated core assigned to general-purpose tasks to do prefectching of a page table into shared physical cache for a soft real-time task. As the page table is referenced frequently by tasks, PTP gives a considerable advantage because fewer TLB misses occur. The results from the Cyclitest and CPUspec2006 benchmarks show that both CAP and PTP can provide better response for real-time tasks. CAP has benefits due to its simplicity, but PTP has advantages in flexibility and high cache utilization due to its dynamic cached maintenance. In addition, the experimental results show that PTP improves real-time performance more but sacrifices some GP performance.

Parallel approaches to machine learning—A comprehensive survey

Literature has always witnessed efforts that make use of parallel algorithms / parallel architecture to improve performance; machine learning space is no exception. In fact, a considerable effort has gone into this area in the past fifteen years. Our report attempts to bring together and consolidate such attempts. It tracks the development in this area since the inception of the idea in 1995, identifies different phases during the time period 1995–2011 and marks important achievements. When it comes to performance enhancement, GPU platforms have carved a special niche for themselves. The strength of these platforms comes from the capability to speed up computations exponentially by way of parallel architecture / programming methods. While it is evident that computationally complex processes like image processing, gaming etc. stand to gain much from parallel architectures; studies suggest that general purpose tasks such as machine learning, graph traversal, and finite state machines are also identified as the parallel applications of the future. Map reduce is another important technique that has evolved during this period and as the literature has it, it has been proved to be an important aid in delivering performance of machine learning algorithms on GPUs. The report summarily presents the path of developments.

Creating portable, repeatable, realistic benchmarks for embedded systems and the challenges thereof

To appreciate the challenges of analysing embedded processor behaviour, step back in time to understand the evolution of embedded processors. Only a few decades ago, embedded processors were relatively simple devices (compared to today), represented by a host of 8- and 16-bit microcontrollers, and 32-bit microprocessors, with minimal integration. Today, these processors (even the so-called, low-end microcontrollers), have evolved into highly-integrated SoCs with a wide variety of architectures capable of tackling both specific and general-purpose tasks. Associated with these transformations, the benchmarks used to quantify the capabilities have also grown in complexity and range. At the simplest level, benchmarks such as CoreMark analyse the fundamental processor cores. At the other end of the spectrum, system benchmarks, such BrowsingBench, analyse the entire SoC as well as the system software stack and even the physical interfaces. This paper examines some of the challenges of applying such benchmarks, and explains the methodologies used at EEMBC to manage portability, repeatability, and realism.

An Autonomous Mobile Robotic System for Surveillance of Indoor Environments

The development of intelligent surveillance systems is an active research area. In this context, mobile and multi-functional robots are generally adopted as means to reduce the environment structuring and the number of devices needed to cover a given area. Nevertheless, the number of different sensors mounted on the robot, and the number of complex tasks related to exploration, monitoring, and surveillance make the design of the overall system extremely challenging. In this paper, we present our autonomous mobile robot for surveillance of indoor environments. We propose a system able to handle autonomously general-purpose tasks and complex surveillance issues simultaneously. It is shown that the proposed robotic surveillance scheme successfully addresses a number of basic problems related to environment mapping, localization and autonomous navigation, as well as surveillance tasks, like scene processing to detect abandoned or removed objects and people detection and following. The feasibility of the approach is demonstrated through experimental tests using a multisensor platform equipped with a monocular camera, a laser scanner, and an RFID device. Real world applications of the proposed system include surveillance of wide areas (e.g. airports and museums) and buildings, and monitoring of safety equipment.

Active Surveillance of Dynamic Environments using a Multi-Agent System

Mobility and multi-functionality have been recognized as being basic requirements for the development of fully automated surveillance systems in realistic scenarios. Nevertheless, problems such as active control of heterogeneous mobile agents, integration of information from fixed and moving sensors for high-level scene interpretation, and mission execution are open. This paper describes recent and current research of the authors concerning the design and implementation of a multi-agent surveillance system, using static cameras and mobile robots. The proposed solution takes advantage of a distributed control architecture that allows the agents to autonomously handle general-purpose tasks, as well as more complex surveillance issues. The various agents can either take decisions and act with some degree of autonomy, or cooperate with each other. This paper presents an overview of the system architecture and of the algorithms involved in developing such an autonomous, multi-agent surveillance system.

Information-seeking behaviours and cognitive search strategies in different search tasks on the WWW

This study investigates the cognitive strategies of 80 participants as they engaged in two researcher-defined tasks and two participant-defined information-seeking tasks using the WWW. Each researcher-defined task and participant-defined task was further divided into a directed search task and a general-purpose browsing task. On the basis of retrospective verbal protocols, log-file data and observations, 12 cognitive search strategies were identified and explained. The differences in cognitive search strategy choice between researcher-defined and participant-defined tasks and between directed search and general-purpose tasks were examined using correspondence analysis. These cognitive search strategies were compared to earlier investigations of search strategies on the WWW. Relevance to industry Describing information-seeking behaviours and cognitive search strategies in detail provides website developers and search engine developers with valuable insights into how users seek (and find) information of value to them. Using this information, website developers might gain some knowledge as to how to best represent the content and navigational properties of websites. Search engine developers might wish to make the search and collection strategies more transparent to users. There are also design implications for the designers of web browsers.

An efficient host/co-processor solution for MPEG-4 audio composition

This paper presents an efficient software host/co-processor architecture for the implementation of MPEG-4 audio composition. The proposed solution is based on a specific partition between general-purpose tasks and DSP-oriented functionality, thus achieving portability, efficient partitioning of the processing resources and memory management.

Intelligent Object Handling by Robot - A Fuzzy Approach

Robot manipulators are required to handle variety of objects and the programmer is required to write explicit program steps to handle the objects depending upon their physical properties. This process is tedious and time consuming. It may become forbidding when robots will be required to perform more general purpose tasks. We propose a methodology to handle the objects in an intelligent manner, where programmer, while defining world coordinates, also states the physical properties of the objects. The intelligence provided with the system, depending upon action required and the object's properties, infers about the way the object is to be handled, thereby reduces the programming burden. A new teach pendent, which helps in defining world coordinates and properties of the objects, is also proposed to realize the methodology.