K.K. BEST PATH RESEARCH
All rights reserved.
Our founders got started in Machine Learning long before the current renewed interest in “AI” and Deep Learning and both wrote and trained their first neural networks for recognizing hand-written digits in the early 90s. Deep Learning is an amazing set of technologies and is often the best choice for a given task. However, sometimes there isn’t enough data, or the deployment target has insufficient power or energy provision for a Deep Learning solution. In such cases, a more traditional machine learning approach may be more suitable. Our team can advise on which technologies are the best for your use case, not limiting the possibilities to “AI” and Deep Learning. Where a Deep Learning solution is likely to be the best approach we can advise on every aspect of the development, from data collection to training and evaluation and deployment. Or we can even do everything for you. And everything in between. An important aspect of Deep Learning deployments is maintainability and continuous evaluation to make sure the model doesn’t become stale due to changes in the inputs over time. The BPR service won’t leave you stranded with a model you can no longer maintain; you have the option to maintain your own models once the initial system has been developed, or you can leave it all to us. Members of the BPR team used to write their own neural network training and inference code but recent advances mean that using popular frameworks like Tensorflow and PyTorch are far more flexible and efficient. We have been working with both tools since their initial releases and use these tools internally for both model training and deployment. We have proven developments in the following areas: Object Detection – examples: detecting receipts on low-contrast backgrounds, traffic sign and traffic light recognition in dashcam recordings Natural Language Processing/Natural Language Understanding – document understanding Optical Character Recognition – Recognizing both hand-written and printed characters in Japanese receipts, Speech Recognition – from simple commands to large vocabulary systems we have extensive knowledge in this area
Each member of our team has many years of experience developing software applications at large companies. Their proven track record of deployed applications ranges from Linux services running in the cloud to embedded applications running on iOS, Android and Raspberry Pi. Our team members’ knowledge and skills is not just limited to machine learning deployments either. A non-exhaustive list of the technologies BPR have extensive experience with is: ROS2 – the go-to open source toolkit for Robotics applications OpenCV – the go-to open source toolkit for Computer Vision tasks The primary programming languages we use for almost all software development are C++ and Python.
Our team can advise on the most suitable data for your task choosing from already existing data sources where available, generating artificial data where possible and appropriate, and annotating the data when necessary. If you already have data we can advise on how it should be annotated, or you can leave everything to our team, who each have several years of experience in annotating large volumes of both image and text data.