AI-Machine Learning

Artificial Intelligence, Machine Learning, Deep Learning

Intelligent Learning Machines (ILM) is building increasingly adaptive and intelligent embedded computing systems using machine learning and data mining techniques. Inevitably, the mobile nature of these embedded systems constrains the amount of computing power available. ILM resorts to efficient and highly tuned system designs, such as those featuring convolutional neural networks (CNN) and other type of non-linear transfer functions. ILM’s rooted experience in this domain can help your design or product excel in the marketplace. Examples of such products are vision processing systems, networked command & control systems, safety watchdogs, security systems and cybersecurity monitors.

ILM reference designs include:

• Embedded computer vision processing algorithms and hardware that learn how to recognize various objects or people
  The ability of computers, to see what humans see (or more advanced vision for extra-sensory perception) and act upon the information in a way that is consistent with societal norms, delivers the magic of computers that futurists such as Isaac Asimov have long written about. Much of this functionality rests with object and person detection (OPD). OPD enables computers to take over large swaths of the 4Ds of human task automation (Dangerous, Dirty, Diurnally-opposite, Dreary) that people are highly motivated relegate to machines. OPD can be a computing intensive process, unless it is properly optimized both from software and hardware standpoints. Embedded systems are particularly vulnerable to performance degradation and the associated safety issues, if not designed properly.


• Networked command and control systems that blend classic feedback-control systems with deep learning capabilities
  The intersection of embedded computing platforms and networked machine learning represents a sweet spot for ILM’s customers that seek to supplement or replace a human presence with cost-effective automated systems that can deliver similar results. The ability to learn how to respond to changing conditions surrounding an embedded system has immediate benefits for an individually deployed embedded system. Data mining these conditions and storing them in a cloud-based command & control system, for broadcast to other similar or future embedded systems on the network, helps customers propagate deep learning training information and rare edge cases to their fleet of embedded systems, even if such systems are dissimilar in their detailed design.


• Adaptive safety watchdogs that learn how to recognize useful real-world dangers and take action to mitigate
  Adaptive safety watchdogs (ASW) are starting to emerge in safety domain electronic control units (ECU) present in nearly every safety critical application, such as passenger cars or medical life support devices. Safety watchdogs are extensively used in vetted functionally safe designs. While classic watchdogs are fairly common in today’s ECUs, ASWs are still relatively rare, because of rapidly evolving standards around non-deterministic computing, some of which ILM is actively developing. Indeed, the safety case around adaptive watchdogs requires more complex analysis methodology that still prevents ASW deployment in many applications where life and limb is at stake. ILM has solved many of such questions with designs that can withstand intense safety assessment and audit processes, while delivering the much-needed flexibility of ASWs.


• Cyber security monitors that detect threats before full-blown intrusion
  Machine learning allows cybersecurity solutions to adapt to unforeseen threats against a networked embedded system. See ILM’s cybersecurity offerings for more information.

We provide support to following applications & services, that are based on computer vision technology:

• Multiple Object Detection: classification and localization of different objects in the scene


• Multiple Object Tracking: generating trajectories of objects over time for motion analysis


• Smart Parking Assistance Via Cameras
  • Person Re-Identification: finding other instances of already seen persons across different cameras
  • Long-Term Tracking: generating long-term trajectories of objects across different cameras using re-identification technique
  • Gate Counting: counting the number of people entering/exiting gates -and/or- regions of an arbitrary shape defined by users


• Facial Recognition for Biometric Authentication: human pose estimation and tracking for detailed human body motion analysis


• Content-Based Image Retrieval: finding all images in a set with specific content

Get Started Today

We‘re here to help! Fill out the form below to contact us and we‘ll respond promptly.