TECH OFFERS

Machine Learning (ML) is a sub-field of Artificial Intelligence (AI) where a machine is able to learn without being explicitly programmed. However, before a machine can effectively perform even the simplest AI tasks, e.g. differentiating between images containing an elephant or a tiger, it has to be trained on images containing both animals. To be useful in supervised learning, training data needs to be properly labelled or annotated by a human for the machine to extract the relevant features and produce an ML model that serves its intended purpose. This highlights the important role that data annotation plays in producing robust, accurate ML algorithms in video analytics, natural language processing, and audio processing. However, many organisations that want to embark on their supervised learning journey often face difficulties gaining access to high-quality labelled datasets, known as ground truth data, due to the abundance of low-quality, expensive and unstructured data. This technology offer is a mobile application-based data platform that enables companies to obtain high-quality annotated data. It de-centralises data collection and data annotation tasks into manageable bite-sized chunks for optimal annotation performance and crowd/out-sources the annotation task to a pool of data taggers via a mobile application. Labelling quality is established through a gamification system and a series of built-in verification procedures, including AI-assisted pre-filtering and collective human quality control.

With the breakthrough in 3D vision-based localisation technologies under GPS-incapable environment and expertise in sensor fusion, this technology offer presents promising sensing solution for advanced robotics and automation in Industry 4.0. This solution is built upon the proprietary vision-based sensor with edge computing capabilities, providing best calibration algorithm and controls with the deployment of the multiple sensor technology. The sensing technology enables robots to: Navigate under GPS-incapable environment, across indoors and outdoors spaces Vision-based self-localisation and moving object tracking Object identification and volumetric measurements in 3D For Industry 4.0, these are the areas can be supported: 3D visual data acquisition and simple onboard processing, translating only required information to the main processing board Smart CCTV monitoring & image / object identification Depth perception and measurements for logistics The technology owner is currently looking for operators that would like to transform their business operations, through the use of autonomous robots, and also to achieve the best effects of encapsulating a whole visual monitoring system to optimise their business operations on the ground. Besides, the technology owner is also looking to work with system integrators (eg. Robot makers) to adopt and integrate the technology. The system is fully customizable and configurable based on user’s requirements.

There has been an increasing interest in LED communications, or Light Fidelity (Li-Fi), for secure, wireless communications. Light Emitting Diodes (LEDs) are ubiquitous in our environment and can be used for communications, besides illumination. Data is transmitted by switching the LEDs on and off so rapidly that the eye cannot detect the switching. This opens up a wide range of unused, unregulated spectrum in the visible to infrared wavelengths, which can help address the communications bandwidth crunch. Li-Fi can be used in areas where RF signals cannot be received due to interference or restriction. The highly secure beam can protect sensitive data from cyberattacks. This technology offer is a system for Li-Fi communications, including optical transceivers and dual purpose LEDs for detection and emission. By alternating between positive and reverse bias, a single LED chip can function as emitter and detector. This can reduce the footprint of the optical transceiver for miniature integration in laptop and handphone. The development team has also developed LED communication prototypes for real-time monitoring of sensor readings over a long range of more than 70m and for real-time streaming of high-definition video streaming with low latency. Some application areas include underwater communications, hospitals, military and underground tunnels. The development team is looking for potential end-users of technology and collaboration in technology development.

Due to the high resource costs (data, computational power) associated with the creation of trained neural network models and the widespread application of deep learning in a plethora of sectors/industries, ranging from mobile apps to autonomous driving, trained models are often viewed as Intellectual Property (IP) of the entity that created them. Hence, it is increasingly critical for stakeholders to mark their ownership and protect their models against potential IP infringement. One way to claim ownership is through conventional digital watermarking, however, this technique is susceptible to model extraction attacks and while watermarking a model does not prevent theft, it enables legitimate owners to verify their ownership over stolen assets. This technology offer is a robust watermarking mechanism that protects the ownership of a high-performance neural network model to the entity that has invested resources to facilitate its training and performance tuning. It turns well-known defects of neural networks into a mechanism for verifiable proof of ownership, whenever required. In this instance, backdoors, which are inserted during a model's training phase to intentionally generate erroneous outputs, and adversarial samples (specifically structured perturbations that are entirely unobservable by the human eye), which are able to fool well-trained and high-performing models into misclassifying input data, are used.

Sounds in industries can indicate and even forewarn various critical events, from changes in the processes to developing machine issues. Combining AI, acoustic sensors, and digital signal processing this system aims to detect and predict issues in industries. With this approach it is able to non-invasively convert assets into smart assets with minimal risk and friction. With applications ranging from fault detection and predictive maintenance to more niche needs such as cavitation or fouling detection, the AI can be customized to adapt to the particular facilities’ needs.

Gas leakage detection system is an importance part of a safety system, providing early warning against possible disasters of gas leakage. Gas is increasinly used in both industrial and residential environment, and hazardous gas leakage if goes it go undetected, can be disastous. However, there are not many detectors avaiable in the market to detect hazardous gas leakage due to chemcials or fuels. This technology provides an imaging solution for gas leakage detection of hazarous gas like methane. The technology consists of 2 types of low pixel count infrared (IR) cameras with both advanced sensing techniques and a proprietary algorithm. This allows for: Passive camera: imaging with external IR light sources (sun, heat sources, lamp or other artificial IR light sources etc.). Passive/Active camera: In addition to passive camera, the active mode relies on active IR laser patterns. Similar to wide-field IR cameras, the technology could monitor a larger site and hence, the site monitoring could require fewer detectors and hence, the cost of using conventional cameras as detectors.  The technology allows for hybrid serial and parallel detection with propriety algorithm that uses enhanced frame rate to achieve real-time industrial standard images.

With the prevalence of Internet of Things (IoTs), urban cities have evolved into Smart Cities, where information is constantly collected and fed into the cloud. Hence, there is a need for a smart grid system that can analyse these data and use it to provide real-time or predictive assistance to citizens and businesses to improve quality of life. This technology offer is a highly customisable AI system for real-time monitoring, anomaly detection and predictive maintenance. The AI is able to provide real-time monitoring of desired systems from distribution networks (power, water, and gas) to other applications such as fibre optics cable monitoring. This monitoring system includes an automated detection and localisation of network anomalies and defects based on incoming sensor data. Using data-driven forecast, it is able to provide insights for predictive maintenance and minimise the disruption of the monitored system. The use of this technology is suitable for other cases such as route optimisation, smart mobility and CO2 forecasting. The technology owner is interested in pilot projects, R&D collaboration as well as to license this technology.

An AI-based Optical Character Recognition (OCR) engine has been developed that is able to auto-extract any content from papers, handwritten and printed documents alike, and then directly transporting it to CSV or to internal ERP system. It will first auto-sort the document into each of its type, and extract the content of each individual document entirely, eliminating most of data entry works. It offers high recognition accuracy even for difficult handwritten data. Even though AI will still need human supervision, the system is able to reduce above 80% data entry cost across sectors, such as business process outsourcing (BPO), banking, manufacturing, logistics, retail, ionsurance among others.  For many applications, the technology can reduce burden in manpower, lowers inaccuracy, and promotes digitalisation. The technology owner is looking for partners and users alike, and offer both trial and subscription programs, on-cloud as well as on-premises. Free trial account is available for potential partners or users to validate the effective of the technology and system to extract free-text content from their documents.

Sustainability in beauty is increasingly becoming a priority across the globe. The green consumerism trend is driving more cosmetics companies to seek natural, clean and environmentally friendly ingredients, packaging and green products. This expectation is pushing beauty brands towards new sustainable innovations, such as waterless beauty, upcycling, carbon neutrality and reusable or refillable packaging solutions to meet the demands from the environmentally conscious generation. This has been further emphasized due to the current pandemic situation.   The technology aims to create a moisture-rich cosmetic in waterless form. This revolutionary concept enables personalization of formulations to produce moisture-rich powder that can be transformed into a spreading cream and/or serum easily upon application. The waterless formulation can be readily customised and adapted to suit the specific needs of an identified target market. Due to its waterless nature, the downstream packaging of products comprising this formulation will involve little use of plastics and can be filled into many different types of packaging including recyclable paper, glass and aluminum packaging, thereby reducing carbon footprint.  The technology provider is seeking codevelopment partners and potential licensees who have interests in leveraging on this waterless technology for supporting sustainability in personal care.