Machine learning is a powerful means for rapidly building smart sensing applications using algorithms taught from training data. Naturally, high quality labeled datasets are pivotal to such methods. Anything less invariably leads to lower algorithm performance.
The most common challenge users encounter using AI/ML in development of sensor algorithms is poor underlying train/test data… and for good reason. Nearly all AI frameworks and automated AI development tools leave this task entirely to the user to figure out, providing rudimentary data collection facilities at best.
SensiML differs considerably and has you covered with equal emphasis on supporting the collection of high quality labeled data as well as supporting leading edge AutoML algorithm generation. No other AI tool suite provides the same level of support for front-end data collection and labeling.
To enable high-quality data, SensiML’s Data Capture Lab application was expressly built to enable the collection, cleansing, labeling, and management of train and test datasets using a rigorous workflow supporting one to many users.
Our Data Capture Lab is a full-fledged data collection tool that brings a level of automated dataset management developers are accustomed to seeing in coding IDEs but not from AI tools until now. Features such as visual segment editing, auto labeling, multi-session labeling, multi-user data collection, versioning, and collaborative project dataset management are all built in and easy to implement.
Your data is the most valuable asset in the AI/ML development process, so ensure it’s value will endure.
Our approach focuses on allowing you to build datasets as IP you can fully maintain, modify, explore, extend, and export as your needs evolve. Collecting and labeling train/test data represents the single greatest development expense and source of differentiation in AI/ML algorithm development.
Our Data Capture Lab reflects this importance by allowing users to curate projects over time, annotate with video, audio, and domain expert labels and metadata so data makes sense down the road when viewed by others. Data collected for one platform can be migrated to others and exported and imported to other tools.
It’s your data, so make sure you start your AI project off right by choosing tools that support your need to protect and retain your data investment over time.
* 1Msps sample rates provided on selected hardware platforms.
See how to open SensiML project files using the Data Capture Lab and upload locally stored projects to the SensiML cloud for analysis and collaboration with other project team members.
SensiML project files include all of the raw sensor data, labels, metadata, and configuration settings associated with your edge AI application.
Learn how Data Capture Lab configures a device and its sensor settings, discovers and connects wirelessly to a device for data recording, and streams real-time sensor data into the application.
Recorded raw sensor data is not complete without associated event class labels and metadata. See how these are easily defined within the tool and selected for each capture file as well.
With sensor data collected into individual capture files, it is necessary to define segments of interest and apply label data to establish ground truth for AI train/test in SensiML Analytics Studio.
Data Capture Lab provides a rich UI that makes labeling straightforward, consistent, and reliable. See the details including how labeling can be automated and multiple label sessions allow for experimentation without corrupting existing labels.