東芝情報システム Product Lineup

Do you have any of the following concerns? ● Processing sensor data and 3D data takes time ● Improving prediction accuracy slows down processing ● AI cannot be used in devices that require real-time performance ▼ Leave it to our ultra-fast, high-precision AI prediction technology "RealtimeForesight"! ―― Features ――――――――――――――――――――――― ◆ Our unique algorithm achieves both fast processing and high-precision predictions ◆ Small-scale AI model capable of real-time processing ◆ Easy implementation on edge devices and FPGAs ――――――――――――――――――――――――――――― RealtimeForesight is a technology that predicts 3D data that changes over time using AI technology.

Do you have any concerns about the performance and cost of DNN model development? ● Increasing processing speed reduces inference accuracy ● No effect of acceleration in the target environment (edge devices) ● Repeated manual model modifications, measurements, and evaluations lead to increased labor and time ● Difficult to optimize for embedded devices with limited memory 　　　　　　　　　　　　　▼ Leave it to our "DNN Acceleration Service" to shorten processing time while maintaining accuracy with our unique neural architecture search technology! ――― Features ―――――――――――――――――――― ◆ Accelerate while maintaining accuracy ◆ Provide optimal models in a short period ◆ Optimized for embedded and edge applications ―――――――――――――――――――――――――― The DNN Acceleration Service generates numerous modified models from your DNN model and searches for a DNN model that meets accuracy and performance requirements. By utilizing our unique automation and prediction technologies, we can explore the optimal DNN model in a short time. *DNN: Deep Neural Network*

Do you have challenges with "creating training data" in the fields of embedded systems, automotive, and robotics? ● Creating training data is time-consuming and costly, hindering development ● Manual annotation leads to variability in quality ● A large amount of training data is needed, but cannot be handled manually ● You want to accelerate PoC and validation, but data preparation is a bottleneck 　　　　　　　　　　　▼ Leave it to our automatic annotation service! ―― Features ――――――――――――――――――――――― ◆ Our unique AI technology automatically generates high-precision training data from a small number of training examples ◆ Supports semantic segmentation and bounding boxes ◆ Creates a large volume of training data in a short time, significantly shortening AI development periods ――――――――――――――――――――――――――――― The automatic annotation service is an engineering service that provides a large volume of high-precision training data for image annotation in a short period of time, using tools based on Toshiba Information Systems' unique AI technology, with a small dataset of learning data and target image data provided according to your annotation specifications.

Do you have any of the following concerns? - It takes time to analyze the causes even after detecting waveform anomalies. - Judgments and responses vary depending on the evaluator. - The transfer of technology to younger engineers is not progressing. - I want to reduce the man-hours spent on evaluation tasks. - I want to utilize AI to improve quality. Waveform Insight will solve these issues! Waveform Insight is a tool that works in conjunction with the waveform judgment AI "Waveform Meister" to provide insights on abnormal waveforms: ✓ What is abnormal? ✓ What could be the potential causes? ✓ What responses are effective? It goes beyond simple pass/fail judgments and supports cause analysis and countermeasure considerations, thereby assisting engineers in their decision-making.

In systems that require real-time performance, such as depth sensors, VR, and touch panels, achieving a balance between "processing delay" and "prediction accuracy" is a significant challenge. RealtimeForesight is a next-generation prediction technology that addresses this challenge through our proprietary AI algorithm. It predicts time-series 3D data quickly and accurately, achieving reduced latency, noise reduction, and enhanced feature sharpness. It is also suitable for implementation on edge devices and FPGAs, allowing for deployment across various industrial sectors. Do you face any of the following challenges? - Want to reduce processing delays of sensor data - Want to improve processing speed while maintaining AI prediction accuracy - Looking for a lightweight algorithm that can be embedded as edge AI - Want to reduce noise from ToF sensors or depth sensors - Want to implement AI technology on FPGAs - Want to improve the response performance of systems that require real-time performance ■ Features of RealtimeForesight that support solving these challenges ✓ Proprietary algorithm that achieves both high-speed processing and high-accuracy prediction ✓ Reduced latency through real-time prediction ✓ Lightweight design suitable for edge device and FPGA implementation ✓ Noise reduction and improved recognition accuracy

■Do you have any of the following challenges? - The processing time of the DNN model is long, making real-time processing difficult. - I want to lightweight and speed up the model for implementation on edge devices. - I want to avoid a decrease in image recognition accuracy due to acceleration. - A lot of effort is required for model modification, measurement, and evaluation. - I feel limited in manually searching for the optimal DNN model. - I want to shorten the AI development period and accelerate the market launch of products and services. - I cannot find a model that balances the desired recognition accuracy and processing performance. Our DNN acceleration service supports the acceleration of image-dependent processing while maintaining the accuracy of DNN models. ■Features of the DNN Acceleration Service ✓ Accelerates DNN models while maintaining recognition accuracy. ✓ Automatically generates and selects numerous DNN models. ✓ Reduces the development time and effort required for model improvement.

■Do you have any of these concerns? - It takes a long time to create training data for AI development. - Struggling with a shortage of manpower for annotation tasks. - High outsourcing costs make it difficult to create large datasets. - Want to efficiently create data for semantic segmentation. - Want to reduce the workload for object detection (bounding boxes). - There is a shortage of training data for AI in autonomous driving and manufacturing. - Want to prepare the training data necessary for improving AI model accuracy in a short period. - Want to shorten the duration of AI development projects. Our automatic annotation service can solve these problems! This service takes your training specifications and the target image data for training, and automatically creates high-precision training data using AI. ■Features of the Automatic Annotation Service ✓ AI automatically generates training data. ✓ Supports semantic segmentation and object detection. ✓ Capable of high-precision and large-scale creation. ✓ Shortens AI development time.

Are there any challenges related to "optimization and decision-making" in systems within the embedded, automotive, and robotics fields? ● The combinations are complex, making it impossible to provide optimal decisions in real-time. ● Calculations take time, leading to control delays and performance degradation. ● You want to perform high-precision optimization, but cannot use servers or the cloud. ● In embedded environments, optimization processes are heavy, leading to abandonment of implementation. 　　　　　　　　　　　　　▼ Leave it to our combination optimization solution for embedded systems! ――Features――――――――――――――――――― ◆ Rapid calculation of good solutions for combinatorial optimization problems using quantum-inspired optimization technology. ◆ Contributes to speeding up control and decision-making through low-latency, real-time processing. ◆ Compatible with various hardware such as FPGA and GPU, making it implementable in embedded environments. ――――――――――――――――――――――― This is a combination optimization solution for embedded devices utilizing Toshiba's unique quantum-inspired optimization technology. It can process combinatorial optimization problems such as allocation, route exploration, and matching at the edge, quickly and securely, in embedded devices like automotive equipment, industrial machinery, and robots.

The combinatorial optimization problems in embedded systems are important issues that also affect control accuracy. Toshiba Information Systems provides combinatorial optimization solutions for embedded systems utilizing Toshiba's unique quantum-inspired optimization technology. We quickly compute good solutions to combinatorial optimization problems and support parallel processing on diverse hardware. Furthermore, by processing in an embedded environment, we assist in achieving a low-latency and secure environment. ■ Do you have these challenges? - Want to process combinatorial optimization problems quickly - Want to perform optimization processing with limited resources of embedded devices - Face strict hardware constraints and cannot achieve sufficient performance - Want to realize processing that leads to real-time decision-making and control - Want to process without relying on an online connection environment - Want to achieve a low-latency processing environment in embedded systems - Want to perform optimization processing in a secure environment - Considering optimization processing using FPGAs and other technologies ■ We propose our combinatorial optimization solutions for embedded systems to address these challenges! ✓ Quickly compute good solutions to combinatorial optimization problems ✓ Capable of parallel processing on diverse hardware ✓ Achieve a low-latency and secure environment