iAbra
Embedded AI Image Analysis
Founded in 2010, iAbra’s focus is deriving utility from the ever growing amounts of sensor data in the world.
Artificial Intelligence is moving from the cloud to embedded applications. This is enabling a wide range of use cases including automotive, defence, manufacturing, healthcare and agriculture.
Machine Learning ecosystems have been built mostly on GPUs for the cloud, where neural networks are generally very large and inefficient. These cloud based neural networks and power hungry GPUs make the AI use cases requiring embedded inference more difficult or impossible to engineer. The FPGA’s efficiency and low power consumption are better able to deliver embedded applications.
iAbra’s tools ensure that efficient neural networks are created and optimised for embedded FPGA silicon, delivering low power AI inference for a variety of applications.
AI Tool Chain
iAbra’s Tools provide an Ecosystem to Target FPGA Silicon for embedded AI Inference.
Targeting FPGA for Neural Network Inference, or embedded AI, requires optimisation at the neural network training stage and an inference architecture that exploits the inherent advantages available to the FPGA silicon. These advantages are most valuable in embedded solutions where power, heat, silicon footprint and environmental harshness are constraint factors.
iAbra Neural PathWorks
iAbra Neural PathWorks is the machine learning software platform from iAbra. It provides the neural network creation element of our end to end tool chain for embedded AI inference on FPGA.
PathWorks provides the following elements:
User Interface
A web based user interface ensures both ease of access and an intuitive process. The interface abstracts the complexity of the required data and computer science to create a unique neural network from data.
Data Preparation.
The data used to train an FPGA specific neural network is prepared within PathWorks, enabling the high signal to noise masking and annotation necessary.
Evolutionary Architecture
Discovering the optimal neural network structure and associated parameters for specific AI tasks need not be a stochastic process consuming human resource. iAbra leverages scalable High Performance Computing infrastructure and the underlying principles of machine learning to automate the process.
HPC Integration
Machine Learning is a High Performance Computing problem. PathWorks provides the integration to and management of an integrated system negating the need for additional hardware tools.
Machine Learning Algorithm
The trained neural network is required to be small and highly efficient in order to fit within the FPGA fabric. iAbra’s learning algorithm targets specific FPGA silicon for inference.
Precompiled Output.
The neural networks created by PathWorks can be compared, and those with the greatest efficacy selected to be compiled to FPGA hardware for an embedded design.
iAbra Neural Synapse
Neural Synapse is the AI Inference architecture for neural networks created in PathWorks.
An FPGA is ideally suited to AI inference by virtue of how data is processed as a pipeline of functions. Neural Synapse fully exploits the FPGA to contain data ingest, processing and output tasks. It avoids the use of any external memory leading to a very low size, weight and power solution.
iAbra provides hardware development kits to enable the end to end implementation of embedded AI.
FPGA Silicon
FPGAs provide the ideal AI inference platform where low size, weight and power are design criteria. The use of FPGAs in harsh environments where security and ultra low failure rates are required is well proven, while the range of FPGA products allows the optimal logic capacity to be selected for an application.
Training a neural network using GPUs and inefficient architectures tends to result in AI models that are too big and complex for embedded systems where power and silicon resources are constrained. By optimising the neural network at the training stage, iAbra’s Neural PathWorks ensures the network is of minimal size for the problem domain provided in the training set, requiring fewer computational operations for inference. This optimisation effort ensures the network can be stored in the FPGA’s memory and logic elements.
iAbra’s Neural Synapse exploits the flexibility of the FPGA hardware allowing image data requiring analysis by the AI inference model to be pipelined from input to output without the need for additional memory resources . With the performance of the FPGA’s dedicated computing architecture we can reach an optimal solution for various applications.
Working with iAbra
iAbra’s tools enable the creation of neural networks and their implementation on FPGA silicon for AI based image analysis.
PathWorks can be made available on a Software as a Service basis, or licenced for implementation on a customer’s infrastructure.
Synapse can be implemented on a Hardware Development Kit for testing purposes.
iAbra can support customers with a range of services, from training and support through to neural network creation and testing.
Automotive
The rapid evolution from automatic driver assistance systems to the ambition of level 5 autonomy has seen the automotive sector lead much of the recent development in AI based vision systems.
While data centre technologies such as GPUs have been exploited to create neural network models that could interpret the driver’s world, these power hungry systems are poorly suited to embedded AI Inference in the vehicle.
The embedded AI systems for automotive are required to be low power, standards compliant and exhibit ultra low failure rates. iAbra’s tools enable the automotive sector to create their own valuable neural network models, while deploying them on FPGA silicon that is better suited to meet their design criteria.
Defence & Security
The increase in image sensor data in defense and security has resulted in many images and hours of full motion video going unseen. The application of human resource alone looking for needles in this ever growing haystack is unlikely to improve outcomes. This industrial, manual process requires the application of machine learning and artificial intelligence.
Augmentation and amplification of the human analysis effort is required. iAbra’s tools abstract the complexity of machine learning technologies, providing data preparation, computational effort, model creation and integration to FPGA silicon for inference.
Our tools allow analysts to focus on their outcomes rather than data and computer science. iAbra’s embedded AI approach is fundamental where network bandwidth is constrained or denied, which is common where sensors are deployed in remote or hostile environments. The FPGA implementation allows design constraints such as size, weight and power to be met while complying with stringent approvals necessary in this sector.
Automotive
The rapid evolution from automatic driver assistance systems to the ambition of level 5 autonomy has seen the automotive sector lead much of the recent development in AI based vision systems.
While data centre technologies such as GPUs have been exploited to create neural network models that could interpret the driver’s world, these power hungry systems are poorly suited to embedded AI Inference in the vehicle.
The embedded AI systems for automotive are required to be low power, standards compliant and exhibit ultra low failure rates. iAbra’s tools enable the automotive sector to create their own valuable neural network models, while deploying them on FPGA silicon that is better suited to meet their design criteria.
Defence & Security
The increase in image sensor data in defense and security has resulted in many images and hours of full motion video going unseen. The application of human resource alone looking for needles in this ever growing haystack is unlikely to improve outcomes. This industrial, manual process requires the application of machine learning and artificial intelligence.
Augmentation and amplification of the human analysis effort is required. iAbra’s tools abstract the complexity of machine learning technologies, providing data preparation, computational effort, model creation and integration to FPGA silicon for inference.
Our tools allow analysts to focus on their outcomes rather than data and computer science. iAbra’s embedded AI approach is fundamental where network bandwidth is constrained or denied, which is common where sensors are deployed in remote or hostile environments. The FPGA implementation allows design constraints such as size, weight and power to be met while complying with stringent approvals necessary in this sector.
Need to safeguard drones and robotic cars against cyber attacks
Researchers executed successful stealth attacks on real and simulated robotic vehicles,…
read moreSatellite broken? Smart satellites to the rescue
Scientists are developing robotic networks that can work independently but collaboratively…
read moreMilestone in quantum standardization
Researchers have developed a method that could pave the way to…
read more