Page Not Found
Recently, a team from MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL) has brought breakthroughs into modern technology. They have introduced a method that harnesses the power of multiple artificial intelligence (AI) systems to engage in discussions and debates, aiming to reach the most optimal solution for a given question. This approach empowers these advanced language models to enhance their commitment to factual information and improve their decision-making.
The main challenge associated with large language models (LLMs) is the inconsistency in the responses they generate, which can lead to potential inaccuracies and flawed reasoning. This novel strategy allows each (AI) agent to actively evaluate the reactions of every other agent and use this collective feedback to refine its response.
Technically, this process includes multiple rounds of response generation and critique, with each language model updating its answer based on feedback from other agents. It culminates in a final output through a majority vote, akin to a group discussion where participants collaborate to reach a unified, well-reasoned conclusion.
A significant advantage of this approach is its easy application to existing black-box models, specifically large language models (LLMs). It smoothly integrates with them, focusing on text generation, and doesn’t necessitate access to their internal workings. This simplicity can make it more accessible for researchers and developers to improve the accuracy and consistency of language model outputs.
Yilun Du, an MIT PhD student in electrical engineering and computer science and an MIT CSAIL affiliate, states, “Rather than relying solely on a single AI model for answers, our process engages a multitude of AI models, each offering unique insights to address a question.
Though initial responses may be brief or contain errors, these models improve by analysing peers’ responses, enhancing problem-solving skills, and validating accuracy through dialogue. It contrasts with isolated AI models often replicating internet content, fostering more precise solutions.
The study concentrated on math problem-solving, yielding significant performance improvements through the multi-agent debate method. Additionally, language models exhibited improved arithmetic skills, suggesting potential applications across various domains.
Furthermore, this method can help address the issue of “hallucinations” commonly encountered in language models. By creating an environment where agents assess each other’s responses, they are more motivated to avoid generating random information and prioritise factual correctness.
Beyond its relevance to language models, this approach can potentially integrate diverse models with specialised skills. Establishing a decentralised system where multiple agents interact and debate could enable the application of these comprehensive and efficient problem-solving abilities across different modalities, such as speech, video, or text.
While promising, the researchers recognise that current language models may struggle with lengthy contexts and that critique capabilities need refinement. The multi-agent debate format, inspired by human group interactions, has room for further exploration in complex discussions crucial for collective decision-making. Advancing this technique may require a deeper understanding of the computational foundation.
Yilun Du noted, “This approach not only offers a way to elevate the performance of existing language models but also provides an automatic mechanism for self-improvement. By utilising the debate process as supervised data, language models can enhance their accuracy and reasoning abilities autonomously, reducing their dependence on human feedback and offering a scalable approach to self-improvement.
As researchers continue to refine and explore this approach, we can move closer to a future where language models mimic human-like language and exhibit more systematic and dependable thinking, ushering in a new era of language comprehension and application.”
The University of Michigan has developed machine-learning algorithms technology. This new technology can identify problematic areas in antibodies, making them less susceptible to binding non-target molecules. This innovative development, led by Peter Tessier, the Albert Mattocks Professor of Pharmaceutical Sciences at U-M and the study’s corresponding author in Nature Biomedical Engineering, presents a ground-breaking solution to enhance the effectiveness of antibodies in fighting diseases.
“Antibodies play a crucial role in our immune system’s defence mechanism by binding to specific molecules known as antigens on disease-causing agents, such as the spike protein on the COVID-19 virus,” Tessier expressed, “Once bound, antibodies either directly neutralise harmful viruses or cells or signal the body’s immune cells to take action.”
However, there’s a challenge associated with antibodies designed to bind strongly and rapidly to their specific antigens. These antibodies may also bind to non-antigen molecules, leading to their premature removal from the body. Moreover, they can interact with other antibodies of the same type, forming dense solutions that do not easily flow through the needles used for delivering antibody drugs.
Tessier highlighted the importance of antibodies that can simultaneously perform three critical tasks: tightly binding to their intended target, repelling each other, and disregarding other substances within the body. Antibodies failing to meet all three criteria are unlikely to be successful drugs. Unfortunately, a significant number of clinical-stage antibodies fall short in this regard.
In their recent study, Tessier’s team assessed the activity of 80 clinical-stage antibodies in the laboratory. It made a startling discovery – 75% of these antibodies interacted with the wrong molecules, with each other, or both. To address this issue, the team turned to machine learning.
By making subtle changes to the amino acids that make up an antibody, they can alter the antibody’s three-dimensional structure. This modification helps prevent antibodies from behaving improperly, as an antibody’s structure determines the substances it can bind to. However, making changes without careful consideration can introduce more problems than they solve, and the typical antibody contains hundreds of amino acid positions that could be altered.
Fortunately, machine learning offers a streamlined solution. Tessier’s team created models that are trained using experimental data collected from clinical-stage antibodies. These models can precisely identify how to modify antibodies to ensure they meet all three criteria mentioned earlier, with an impressive accuracy rate of 78% to 88%. This approach significantly reduces the number of antibody modifications that chemical and biomedical engineers need to produce and test in the lab.
Tiexin Wang, a doctoral student in chemical engineering and a co-author of the study, emphasised the pivotal role of machine learning in accelerating drug development. This advanced technology is already attracting attention from biotech companies, which recognise its potential for optimising the development of next-generation therapeutic antibodies.
Tessier concluded by mentioning that some companies have developed antibodies with desirable biological activity but are aware of potential challenges when using these antibodies as drugs. In such cases, Tessier’s team steps in to identify specific areas within the antibodies that require modification, offering valuable assistance to these companies.
The Hong Kong Polytechnic University (PolyU) has secured substantial funding support totalling HK$25.1 million for a selection of 27 innovative projects from the Health and Medical Research Fund (HMRF) during the 2021 funding allocation. This represents a noteworthy increase in both the number of projects and the total funding awarded, underlining the university’s commitment to advancing research in the fields of health and medicine through technology and innovation.
The projects receiving these awards are driven by researchers from various faculties within the university, including the Faculty of Health and Social Sciences, Faculty of Humanities, Faculty of Science, and the School of Design. These projects collectively aim to develop pioneering and impactful solutions geared towards enhancing the quality of medical and mental healthcare services.
The research initiatives encompass an extensive spectrum of topics and age groups, addressing critical health concerns and providing innovative solutions. Some of the projects focus on cognitive and physical training programmes tailored for older adults, early detection of postpartum depression, adolescent idiopathic scoliosis, Parkinson’s disease, autism spectrum disorder, and system-biology analytics for schizophrenia. Other projects aim to advance medical progress in critical areas such as liver cancer radiotherapy and palliative care.
In the realm of eye health, the projects delve into areas such as myopia control, the development of an anti-glaucoma agent, and the creation of drugs for antibiotic and infection treatment. Additionally, several projects concentrate on providing support for caregivers by enhancing their mental health services while also catering to the specific needs of patients.
One notable aspect of this research is the integration of artificial intelligence (AI) technologies, including deep learning and machine learning, to enhance clinical diagnosis and analysis. For instance, Prof Weixong ZHANG, Chair Professor of Bioinformatics and Integrative Genomics at the Department of Health Technology and Informatics, is leading the project “Subtyping and Diagnosis of Schizophrenia by Systems-biology Analytics.” This initiative employs AI technologies to integrate genomic and neuroimaging data, facilitating a deeper understanding of schizophrenia’s etiology and subtypes, with the ultimate goal of enabling personalised medicine for affected individuals.
In the context of liver cancer radiotherapy, Dr Tian LI, Research Assistant Professor of the Department of Health Technology and Informatics, is spearheading the project “Investigation of a Deep Learning-empowered 4D multi-parametric MRI (4D-mpMRI) Technique for Liver Cancer Radiotherapy in a Prospective Clinical Trial.” This endeavour aims to enhance the image quality and clinical effectiveness of 4D-mpMRI radiotherapy techniques by leveraging deep learning, ensuring more accurate measurement of tumour motion and volume for improved treatment planning.
Technology is also playing a pivotal role in the field of rehabilitation management. Autism spectrum disorder, a condition without a cure, presents unique challenges. Dr Yvonne Ming Yee HAN, Associate Professor of the Department of Rehabilitation Sciences, leads the project “Cumulative and Booster Effects of Multisession Prefrontal Transcranial Direct-current Stimulation on Cognitive and Social Impairments in Adolescents with Autism Spectrum Disorder.” This research explores the potential long-term efficacy of transcranial direct current stimulation (tDCS) in mitigating core symptoms in individuals with autism, offering hope for improved cognitive and social functioning.
Creative technologies are also being harnessed to enhance the effectiveness of physical and mental health treatments in both clinical and community settings. Dr Shanshan WANG, Research Assistant Professor of the School of Nursing, leads the project “Effects of e-bibliotherapy on the Psychological Wellbeing of Informal Caregivers of People with Dementia: A Randomized Controlled Trial.” This initiative has developed an e-bibliotherapy app/manual aimed at improving the psychological well-being and health-related quality of life for caregivers of dementia patients.
In an innovative endeavour, Dr Hailiang WANG, Assistant Professor of the School of Design, is leading the project “A VR-based Real-time Interactive Tutoring System for Qigong Training among Older Adults with Mild Cognitive Impairment and Their Familial Caregivers: A Feasibility Study.” This project explores the integration of traditional exercise Qigong with virtual reality (VR)-based training to design a platform that enables older adults to engage in Qigong exercises, potentially delaying the progression of dementia.
PolyU’s securing of funding aligns with Hong Kong’s broader efforts to drive positive societal change through technological innovation, echoing the collaborative spirit seen in initiatives like the “Healthcare Innovation Challenge” organised by the Hong Kong Science and Technology Parks Corporation (HKSTP), as reported earlier by OpenGov Asia, in partnership with a US-based pharmaceutical company. These endeavours collectively demonstrate the region’s dedication to leveraging technology to shape the future of healthcare and advance societal well-being.
A groundbreaking partnership between Monash University and the Australian Federal Police (AFP) resulted in a cutting-edge research centre known as the AI for Law Enforcement and Community Safety Lab (AiLECS). The lab harnesses artificial intelligence (AI) to drive technology-based initiatives that support law enforcement efforts and enhance safety within local and global communities, particularly in the digital realm.
The official launch of AiLECS marked a significant milestone in the realm of AI and law enforcement. The event was graced by AiLECS Co-Directors Associate Professor Campbell Wilson from Monash University and AFP Leading Senior Constable Dr. Janis Dalins, alongside Monash University Interim Vice-Chancellor and the AFP’s Deputy Commissioner.
The AiLECS Co-Director highlighted the transformative impact of emerging technologies on information accessibility and content creation. He underscored that the same technologies that offer tremendous potential for social good can also be misused, leading to a surge in cyber-attacks, identity theft, exploitation, and the proliferation of misinformation.
The research undertaken at AiLECS is at the forefront of leveraging machine learning, natural language processing, network analysis, and other AI techniques to empower law enforcement. Its scope encompasses countering child abuse material, detecting and classifying illegal firearms, identifying misinformation, and analysing expansive online criminal networks. An essential aspect of their work is the ethical sourcing of datasets to ensure that the AI systems they develop are not only effective but also responsible.
AiLECS originally emerged as a research lab in 2019 and has since initiated several projects aimed at bolstering community safety and providing support to law enforcement agencies. Monash University’s Interim President and Vice-Chancellor hailed AiLECS as a beacon of technological expertise leading the way in creating resilient and responsible initiatives. She stressed that these initiatives are pivotal in fostering safer and thriving communities worldwide.
In addition to the Monash University AI and technology scientists, the research centre boasts a diverse team that includes representatives from the AFP and seasoned experts with experience in law enforcement. Notable among them is Professor Jon Rouse APM, renowned for his pioneering work in countering child exploitation and his former leadership of the globally acclaimed ‘Taskforce Argos’ within the Queensland Police Service.
The Deputy Commissioner of the AFP emphasised the critical importance of this collaboration in combating tech-savvy criminals. These individuals increasingly leverage technology to facilitate illegal activities, which pose significant challenges to national security, social harmony, and economic stability. She pointed out that the partnership also aims to address pressing concerns related to privacy, AI, and machine learning to ensure that these technologies are deployed responsibly for the benefit of society.
The other Co-Director of AiLECS and a Leading Senior Constable in the AFP underscored the necessity for law enforcement agencies to actively engage with emerging technologies. He emphasised that through this partnership, AiLECS aims to merge cutting-edge research in AI and machine learning with the principles and expertise of law enforcement, becoming a leading voice for ethics and accountability in AI.
An example of AiLECS’s collaborative efforts is Project Metior Telum. In this project, Monash researchers, in collaboration with the AFP and an industry partner, have harnessed photogrammetry and 3D scanning technology to construct a comprehensive digital library of firearms.
This digital resource enables the rapid development of next-generation tools to detect and combat firearms trafficking. With Metior Telum, every element of the firearm library, from ownership records to specific models, can be meticulously traced.
The AFP, through the Commonwealth Confiscated Assets Account, has extended support to AiLECS Lab activities through a generous four-year funding program. Monash University has also made substantial contributions to this initiative, emphasising the importance of this collaboration in advancing technology and AI for the greater good of law enforcement and community safety.
In a remarkable stride towards embracing AI’s potential, the School of Law at National Yang-Ming Chiao Tung University (NYCU) has introduced a ground-breaking course. This course, titled “AI-Powered Legal Writing: Techniques, Tactics, and Tools,” aims to equip future legal practitioners with the skills needed to harness generative AI in the composition of legal documents.
In recent years, AI has emerged as a transformative force in the legal profession. Reports have indicated that AI can handle a substantial 44% of legal tasks, with administrative functions being particularly vulnerable to automation. Tasks such as researching legal precedents and drafting legal pleadings have already been streamlined and optimised by AI systems.
However, the integration of AI into the legal arena is not without its challenges. In the United States, an attorney faced disciplinary action for employing generative AI to prepare court documents. This incident underscores the importance of law students comprehending both the capabilities and limitations of AI. Failure to do so could result in ethical breaches and threaten their future careers in law.
AI’s influence extends beyond the classroom and into the realm of professional examinations. Associate Professor Mark Shope, from the School of Law at NYCU, astoundingly used generative AI to answer multiple-choice questions on the Taiwan Lawyer’s Bar Exam. While his score of 342 out of 600 fell short of the required threshold of 372 to advance to the exam’s second stage, it surpassed the performance of approximately half of the examinees. This achievement underscores AI’s potential as a valuable tool in the legal profession.
The “AI-Powered Legal Writing” course, spearheaded by Assoc Prof Mark, fills a critical void by focusing on the integration of AI tools into the legal writing process. Assoc Prof Mark emphasises that the adoption of AI is rapidly gaining traction in the realm of legal research and writing. The curriculum not only acquaints students with generative AI but also delves into various generative AI tools and their practical applications in the legal field.
One pivotal aspect emphasised by Assoc Prof Mark is that legal outcomes are often communicated through written documents. AI proves invaluable in composing these legal documents, organising information, and preparing for litigation. Consequently, numerous law firms have already embraced AI to enhance their operational efficiency. It is crucial for aspiring lawyers to recognise that, like professionals in other domains, they face the risk of being replaced if they do not adapt to this evolving technological landscape.
During legal proceedings, legal teams typically expend substantial time and effort analysing cases, studying legal precedents, and reviewing extensive data. AI has the potential to revolutionise this process by reducing the need for a large number of lawyers, thereby increasing access to legal services for the public. These changes in the legal landscape have broader implications for legal education, necessitating a shift in how law students are prepared for their future careers.
The integration of generative AI into legal education offers a myriad of advantages to law students. From enhancing efficiency and research capabilities to providing practical experience and cost savings, AI is a valuable tool that equips aspiring lawyers with the complexities of the modern legal profession. As technology continues to reshape the field of law, students who embrace AI are poised to excel and contribute meaningfully to the ever-evolving legal landscape.
In today’s digital era, the intersection of artificial intelligence (AI) and governance is gaining prominence as nations strive to ensure AI benefits the greater good. Singapore, a firm advocate for AI in the public interest, emphasises the importance of AI governance in the pursuit of a technologically advanced yet ethical society.
Josephine Teo, the Minister for Communications and Information, shared her insights on AI governance for the greater good at the Tallinn Digital Summit 2023, shedding light on Singapore’s approach to harnessing AI for the benefit of society. In her address, Minister Josephine outlined three key objectives that underpin Singapore’s AI governance strategy.
First is the expanding opportunities. Singapore’s digital governance approach aims to create an environment where people and businesses can thrive. This includes substantial investments in digital infrastructure, fostering capability development, and adapting and updating regulations to keep pace with technological advancements.
Second is ensuring trust and safety. Trust is paramount in the digital age. When Singapore introduces new digital laws or updates existing ones, it actively engages with stakeholders, particularly companies affected by the regulations. Building relationships and involving developers in crafting solutions are key components of this approach. While understanding technology deeply is essential, recognising the limits of such understanding is equally crucial.
The third one is strengthening the community. AI governance should strengthen society. Singapore believes in fostering a sense of community and whole-of-society involvement. International collaboration and cooperation are central to achieving this goal, leading to the convergence of principles, rules, systems, and standards.
Addressing the question of whether a global AI agency should be established, Minister Josephine emphasised the importance of international cooperation in the digital domain. She highlighted the effectiveness of international organisations like the International Atomic Energy Agency (IAEA), International Civil Aviation Organisation (ICAO), International Labour Organisation (ILO), and International Maritime Organisation (IMO) in governing global commons.
Key factors contributing to the success of these organisations include clear mandates backed by internationally recognised treaties and conventions, enforceable standards, and sustained evolution to adapt to changing needs. Minister Josephine stressed the need to invest in building the foundation for international consensus on AI governance, citing the Counter Ransomware Initiative (CRI) as a useful reference.
The CRI, a targeted yet comprehensive initiative, focuses on ransomware while covering all aspects related to it. CRI implements common rules, such as preventing ransom monies identified in one country from leaving the financial system of another, thus closing a significant loophole exploited by ransomware criminals. The initiative adopts a pragmatic and inclusive approach to membership, welcoming countries willing to enforce the same rules.
Singapore’s commitment to digital cooperation extends beyond AI governance. The ASEAN Member States have made significant progress in digital cooperation, with initiatives such as a data management framework, Model Contractual Clauses for Cross Border Data Flows by businesses, and discussions on a Digital Economy Framework Agreement.
According to Minister Josephine, Singapore will assume the chairmanship of the ASEAN Digital Ministers Meeting (ADGMin) in January 2024, with plans to launch an ASEAN Guide on AI. She added that these building blocks contribute to the development of global agreements, laying the foundation for a collaborative approach to AI governance.
The Minister stressed that the Tallinn Digital Summit (TDS) serves as a crucial platform for international digital cooperation, addressing the challenges and opportunities presented by the digital era; hence the summit reflects Singapore’s commitment to global technology governance.
The Lee Kong Chian School of Medicine (LKCMedicine) at Nanyang Technological University, Singapore (NTU Singapore), is gearing up for a transformative shift in its curriculum, set to reshape the future of medical education. Starting in 2024, LKCMedicine will embark on an ambitious journey to equip its students with cutting-edge knowledge and skills, blending traditional medical expertise with the transformative powers of artificial intelligence (AI) and digital health.
The LKCMedicine has devised a forward-looking curriculum that aims to produce doctors who are not only compassionate healers but also discerning and confident users of technology. The key pivot in this educational revolution is the integration of digital health and AI throughout the five-year Bachelor of Medicine and Bachelor of Surgery (MBBS) degree programme.
Instead of relegating these critical subjects to isolated modules, LKCMedicine has chosen to infuse them as vertical courses, seamlessly interwoven into the fabric of medical education. This visionary approach empowers students to develop a deep understanding of medical data science, data analytics, and the ethical and legal dimensions of AI in healthcare.
But this transformation goes beyond theory. LKCMedicine is committed to hands-on learning, ensuring that students gain practical exposure to AI and medical technologies such as telehealth, health apps, wearables, and personalised molecular medicine. By doing so, they will graduate not just as doctors but as tech-savvy medical professionals poised to harness the full potential of digital innovations for the betterment of patient care.
To facilitate this bold endeavour, LKCMedicine is introducing an array of tech-enabled learning tools. Virtual reality simulations, like the upcoming virtual reality learning tool for the heart, are poised to revolutionise the way students comprehend complex anatomical systems.
Moreover, the development of custom-built e-simulators for drug prescription and electronic medical record management will mirror real-world hospital settings, ensuring that graduates are well-prepared for the challenges of clinical practice in the digital age.
However, amidst this high-tech revolution, the school remains resolute in preserving the essence of compassionate care. The human touch, which is at the heart of medicine, will not be overshadowed by machines. LKCMedicine recognises that doctors need to be adaptable and agile, capable of navigating uncertainties and delivering patient-centred care.
This commitment to human-centric medicine is exemplified by the expansion of medical humanities in the curriculum. Drawing from the realms of Arts, Humanities, and Social Sciences, this interdisciplinary approach equips students with the skills to manage clinical uncertainty and adapt to the ever-evolving landscape of healthcare.
By integrating medical humanities throughout the five-year programme, LKCMedicine aims to foster adaptability and resilience among its students. In an era when AI and technology are becoming integral to medical practice, medical humanities will serve as a compass, guiding students to keep the patient at the centre of their decision-making process.
As AI becomes a prominent presence in clinical environments, there is a risk that patient-centred care might wane. Medical humanities bridge this gap by equipping future doctors with the critical skills to engage with technology while maintaining their focus on patient well-being.
LKCMedicine’s commitment to holistic medical education extends to its admissions process. Beginning in 2024, the school will replace the BioMedical Admissions Test (BMAT) with the University Clinical Aptitude Test (UCAT).
This shift reflects a broader perspective on what makes a great doctor, assessing qualities beyond academic prowess. Compassion, teamwork, problem-solving abilities, and integrity are now given the prominence they deserve in the selection of future medical professionals.
MIT and a collaborative research initiative have developed an efficient computer vision model that enables autonomous vehicles to rapidly and accurately recognise objects, even in high-resolution images. This model reduces computational complexity, allowing real-time semantic segmentation on devices with limited hardware resources, like those used in autonomous vehicles, for quick decision-making.
Recent cutting-edge semantic segmentation models directly capture pixel interactions in images, causing computational demands to increase exponentially with image resolution, limiting real-time processing on edge devices like sensors or mobile phones.
MIT researchers have introduced a novel building block for semantic segmentation models, matching the capabilities of models but with linear computational complexity and hardware-friendly operations. As a result, this new model series enhances high-resolution computer vision, achieving up to nine times faster performance on mobile devices while maintaining or surpassing accuracy levels.
Beyond aiding real-time decisions in autonomous vehicles, this technique has potential applications to enhance efficiency in other high-resolution computer vision tasks, including medical image segmentation.
According to Song Han, a senior author of the paper and an associate professor in the Department of Electrical Engineering and Computer Science (EECS) at MIT, although traditional vision transformers have been in use for a significant period and yield impressive results, they aim to draw attention to the efficiency dimension of these models. Their research demonstrates the feasibility of significantly reducing computational requirements, enabling real-time image segmentation to occur locally on a device.
Han acknowledged that categorising each pixel in a high-resolution image with potentially millions of pixels poses an intricate challenge for machine learning models. He explained that recently, a highly effective model called a vision transformer has emerged, proving its effectiveness in addressing this challenge.
Initially designed for natural language processing, transformers represent each word in a sentence as a token and then create an attention map to capture the relationships between all permits, facilitating contextual understanding during predictions. Similarly, a vision transformer applies this concept to images by dividing them into patches of pixels and encoding each patch into a token, subsequently generating an attention map.
This attention map relies on a similarity function to directly learn pixel interactions, resulting in a global receptive field that allows the model to access all relevant parts of the image. However, when dealing with high-resolution images comprising millions of pixels organised into thousands of patches, the attention map becomes exceedingly large, leading to quadratic growth in computational demands as image resolution increases.
In their novel model series, known as EfficientViT, the MIT researchers simplified the creation of the attention map by substituting the nonlinear similarity function with a linear one. This alteration allowed them to rearrange the order of operations, reducing the overall computational workload without altering functionality and sacrificing the global receptive field. Consequently, their model exhibits linear growth in computation requirements as image resolution increases.
However, this linear attention approach primarily captures global image context, leading to a decline in accuracy due to the loss of local information. The researchers integrated two additional components into their model to address this accuracy loss, each incurring minimal computational overhead. One of these elements assists in capturing local feature interactions, compensating for the linear function’s limitations in local information extraction.
The second element, a module enabling multiscale learning, facilitates recognising large and small objects. The researchers emphasised the delicate balance between performance and efficiency in their design. EfficientViT is engineered with a hardware-friendly architecture, making it suitable for deployment on various devices, such as virtual reality headsets and edge computers in autonomous vehicles. Moreover, this model can be applied to diverse computer vision tasks, including image classification.
In tests on semantic segmentation datasets, the researchers found their model, EfficientViT, performed up to nine times faster on Nvidia GPUs compared to other popular vision transformer models, maintaining or surpassing accuracy. This advancement enables the model to run efficiently on mobile and cloud devices. The researchers plan to extend this technique to accelerate generative machine-learning models and develop EfficientViT for various vision-related tasks.