Embedded Machine Learning Project? Shorten Your Market Time with an ML Design Team

Are you launching a Machine Learning (ML) solution across embedded platforms? You can speed up your mission with an experienced ML Embedded Design Team.

NOVELIC's ML Design Teams create data acquisition, annotation, training, validation, and model architectures using industry-standard frameworks. Using our cross-disciplined HW, SW, and ML Virtual Design Center teams, we develop automatic acquisition and annotation pipelines.

Our ML Portfolio

Novelic ML Embedded Design Center Team project examples:

Automotive In-Cabin Radar Sensing

NOVELIC's transformative In Cabin Machine Learning software-defined radar track Child Presence, Seat Occupancy, Intrusion & Proximity, Vital Signs, Gestures, and Pet Detection, utilizing Time Distributed Convolutional-Recurrent, Residual Attention, and Vision Transformer neural networks. Our model deployment integrates seamlessly into embedded systems. We balance model efficiency, computational constraints, and uncompromised user experience using meticulous quantization processes and optimization.

Industrial Safety and Comfort Applications

NOVELIC applies ML to enhance workplace safety using software-defined radar to distinguish between moving humans and robotic machines, creating human presence detection zones. Drawing insights from public radar datasets, our team implemented a Radar Multiple-Perspectives (RAMP) Convolutional Neural Network (CNN), incorporating classical radar preprocessing techniques to solve complex industrial real-world challenges.

Gesture Recognition

Our gesture recognition ML project identified and categorized swipes, spins, and specific letter shapes using a range of Neural Network (NN) architectures, including Convolutional, ResNet, LSTM, and Transformer NN models. Applications include touchless infotainment, doors, tailgates, sunroofs, windows, elevators, and intelligent building controls.

Our ML Capabilities

Models and Frameworks

Teams utilize diverse deep-learning architectures within our Computer Vision framework. We harness the core capabilities of CNNs, Residual Networks (ResNets), Long Short-Term Memory (LSTM) networks, Attention-Based Transformer, and mesh architectures like the Residual Attention Network (RAN). We use industry-standard frameworks such as TensorFlow and PyTorch. These tools provide the flexibility and efficiency required to deploy sophisticated ML models.

Data Acquisition and Training Pipelines

Our teams use commercially available and bespoke application-specific data sets. We are Sensor Fusion data acquisition experts with years of experience synthesizing Vision-Radar systems. Our data collection, preprocessing, and ML model training utilizing predefined optimal machine learning models autonomously validate uploaded data, initiate preprocessing without developer intervention, and train the model using the enriched training set.

Operationalizing ML

Our proprietary model tuning tools and hyperparameter optimization under limited acceleration and memory ensure optimal performance. Our model quantization experience extends the functionality of TensorFlow's model optimization toolkit, further enhancing our ability to deploy efficient models on embedded platforms. By automating individual machine-learning models tailored for embedded C, we've streamlined porting models to embedded software.

Contact me for more information--we have an eight-engineer multidisciplined ML team wrapping a project and needing another.

Keywords: #automotiveradar, #NeuralNetworks, #NN, #ConvolutionalNN, Time Distributed Models, Recurrent Network, Residual Attention Network, Vision Transformers, Embedded System Optimization, Object Detection and Classification, Public Radar Dataset, RAMP-CNN, Radar Signal Processing, ResNet, LSTM, Transformers, Automotive Infotainment, Smart houses, Data management, Model optimization

?