Qeexo AutoML 1.16.3

April 26, 2022

New Features:

  • Exposed Arm Virtual Hardware Live Replay (Live Testing) feature for AVH projects

Bug Fixes & Improvements:

  • Fixed segment display issue when navigating away or refreshing segment editor

  • Exposed 1.3KHZ ODR option for STWin low power accelerometer

Known Issues:

  • UMAP / PCA plots visualizations in training timeouts

  • Excessive classes / classes labels may cause issues with live testing

Qeexo AutoML 1.16.2

April 26, 2022

Bug Fixes & Improvements:

  • Fixed issue where segment color was not retained

Qeexo AutoML 1.16.1

April 26, 2022

New Features:

  • Exposed Arm Virtual Hardware M55 MCU device

Bug Fixes & Improvements:

  • Optimized Qeexo AutoML ML pipeline for keyword detection use case

  • Improved spectrogram data segmentation color scheme

Known Issues:

  • Segment color labels do not persist when navigating away from the segment editor 

Qeexo AutoML 1.16.0

April 18, 2022

Bug Fixes & Improvements:

  • Fixed weights / sensitivities apply to device issue under Live Classification Analysis page

  • Fixed error of loading UMAP and PCA plots for grouped datasets

  • Fixed error of UMAP and PCA plot display when visualizing from manual sensor selection during new training process

  • Fixed spectrogram segmenting issue Improved Qeexo Documentation under Help Center

  • Fixed sampling rate mismatching issue for data uploading

  • Improvement of Single Class model’s anomaly detection

  • Improved Training Data Split across ML pipeline

  • Improved Test data evaluation user flow and model performance result display

  • Improved data batching

  • Integrated New UNICO / mlc_configuration_tool Version into AutoML

  • Any model support quantization is now on default

Known Issues:

  • Segment color labels do not persist when navigating away from the segment editor 

Qeexo AutoML 1.15.7

March 16, 2022

Bug Fixes & Improvements:

  • Fixed live test hanging with STWin due to software reset issue

  • Improved data segmentation creation time

  • Fixed issue with spectrogram data segmentation

  • Fixed build failure issue caused by duplicate time-stamps

Known Issues:

  • Segment color labels do not persist when navigating away from the segment editor 

  • UMAP and PCA plot display error when visualizing from manual sensor selection during new training process

Qeexo AutoML 1.15.0

January 14, 2022

New Features:

  • Manual UI Data Segmentation

  • Exposed ‘Current’ sensor for STWin devices

Bug Fixes & Improvements:

  • STWin now supports DFU mode without the need for ST-LINK/v2

  • Improved post-training test data linkage

  • Replaced event data collection option with UI data segmentation

Known Issues:

▪ Segment color labels do not persist when navigating away from the segment editor

Qeexo AutoML 1.14.4

New Features:

  • Support for ST Meta Classifier for MLC projects

  • Edit Label functionality

  • Spectrogram data visualization

  • Arduino BLE classification support for Windows

Bug Fixes & Improvements:

  • Further improved automatic filter & feature selection performance for MLC projects

  • Improved Qeexo AutoML restrictions on ML pipeline input to address build failures

  • Addressed issue with MCU event classification

Known Issues:

  • Data check errors observed on event data collected with Qeexo AutoML

  • Error observed while loading UMAP and PCA plots for grouped datasets

Qeexo AutoML 1.14.3

Bug Fixes & Improvements:

  • Added support for Windows 10 versions 19043.985 and newer

  • Addressed data collection saving issue with large datasets on Windows

  • Addressed issue with event data visualizations for single-channel sensors

Known Issues:

  • Data check errors observed on event data collected with AutoML

Qeexo AutoML 1.14.1

Bug Fixes & Improvements:

  • Added support for data visualization y-axis scroll and y-axis zoom

  • Addressed ‘Device not Found’ error when starting data collection on Windows

  • Addressed ‘Failed to Flash’ issue with STMicroelectronics SensorTile.box and M1 Macs

Known Issues:

  • Limited support for Windows version 1903 and earlier

Qeexo AutoML 1.14.0

October 05, 2021

New Features:

Qeexo AutoML Swift-based desktop client for Windows 10 and macOS

  • Removed network and python dependencies

  • Improved test data management

  • Evaluate test data metrics without retraining models

  • Link training and test data at the label level with exclusion control

  • Improved automatic filter & feature selection performance for MLC projects

  • Added data export functionality

  • Multiple concurrent model builds for PRO tier customers

Bug Fixes & Improvements:

  • Addressed an issue which caused build failures after collection for some single class models on STMicroelectronics SensorTile.box

  • Addressed an issue which caused build failure at calculating features when all models are selected

  • Improved Bluetooth stability during live classification

  • Added support for Apple M1 processor to Qeexo AutoML desktop client

Known Issues:

  • In case that flashing is not successful, resetting the embedded device and retry flashing can help. For Arudino Nano 33 BLE Sense and Arduino Nano 33 IoT, double press the white button to trigger the reset sequence

Qeexo AutoML 1.13.3

Bug Fixes & Improvements

  • Addressed an issue whereby all the candidate features are included in UCF and JSON output for MLC projects. In the release only the final features are exported in those 2 files.

Qeexo AutoML 1.13.2

Bug Fixes & Improvements

  • Added support for UCF file download for MLC projects.

Qeexo AutoML 1.13.1

Bug Fixes & Improvements

  • Addressed an issue to accept appropriate sensor ODR for MLC projects.

Qeexo AutoML 1.13.0

June 30, 2021

New Features

  • Added support for STMicroelectronics Machine Learning Core (MLC) sensors

  • Added a new classification type, “Multi-class Anomaly”, which is a multi-class classification that can also detect an unknown class (“none-of-the-above" scenario)

  • Added FFT power-adaptive binning feature group to the following hardware platforms:
    - Arduino Nano 33 BLE Sense
    - Renesas RA6M3 ML Sensor Module
    - STMicroelectronics SensorTile.box
    - STMicroelectronics STWINKT1B Wireless Industrial Node

Bug Fixes & Improvements

  • Achieved OWASP Application Security Verification Standard (ASVS) Level 1 Compliance (penetration test)

  • Increased data collection time limit to one hour on the following hardware platforms:
    - Arduino Nano 33 BLE Sense
    - Renesas RA6M3 ML Sensor Module
    - STMicroelectronics SensorTile.box
    - STMicroelectronics STWINKT1B Wireless Industrial Node

  • Improved error handling of the CSV data upload process

Known Issues

  • Disabled live classification over Bluetooth for the Arduino Nano 33 BLE Sense and the Arduino Nano 33 IoT on Windows (corresponding Arduino open-source code does not support Windows 10 version 2004 (20H2) or later)

  • Apple computers with M1 chip are currently not supported on Qeexo AutoML (coming soon)

Qeexo AutoML 1.12.0

March 29, 2021

New Features

  • ZMOD4410 indoor air quality sensor and pressure sensor support for Renesas RA6M3 ML Sensor Module

  • Revised PRO and ENTERPRISE tiers with paid subscriptions

  • Single-class classification support for the Arduino Nano 33 IoT (Cortex-M0+)

  • STWIN SensorTile Wireless Industrial Node (STWINKT1B from STMicroelectronics) support

Bug Fixes & Improvements

  • Microsoft Edge browser support

Known Issues

Renesas RA6M3 ML Sensor Module may occasionally stop streaming data, which can affect both data collection and live classification. This issue can be fixed by disconnecting and reconnecting the USB cable from the Renesas RA6M3 to reset device.

Qeexo AutoML 1.11.3

December 16, 2020

New Features

  • Added new machine learning models: Polynomial Support Vector Machine (POLYSVM), RBF Support Vector Machine (RBFSVM), Gaussian Naive Bayes (GNB), One Class Random Forest (ORF).

  • Added optional step to collect and upload Test Data datasets to measure trained model performance against the uploaded datasets.

  • Added PCA plot and F1 score to model details.

  • Added time limit and trial threshold to Hyperparameter Tuning Optimizer.

Bug Fixes & Improvements

  • Removed Chrome extension dependency by adding a Qeexo AutoML native app.

  • STWIN now supports both digital and analog microphones.

  • Added proxy server option for Qeexo AutoML installers to bypass network restrictions.

Known Issues

  • Multiple CSV files upload with different sensor configurations is not supported and may lead to unexpected issues. Each upload should only contain data with the same sensor configurations.

  • Uploading a large amount of data through multiple CSV files at once may fail when server traffic overloads.

Additional Notes

  • SensorTile.Box now supports DFU mode, without the need for ST-LINK/v2.

Qeexo AutoML 1.11.3 (EARLY ACCESS)

November 19, 2020

New Features

  • Qeexo AutoML frontend runs natively, without the need for Chrome extension.

  • Added proxy option for frontend installers.

  • Added Test data management.

  • Added new models: RBF Support Vector Machine (RBFSVM), Gaussian Naive Bayes (GNB), One Class Random Forest (ORF).

  • Added PCA plot and F1 score to model details.

  • Added time limit, trial threshold to Hyperparameter Tuning Optimizer.

Bug Fixes & Improvements

  • STWIN now supports both digital and analog microphones

Known Issues

  • UI issue: RBFSVM stays at “get available memory” stage but the model training is actually complete.

  • AQ-2097

  • OCSVM may fail in large dataset

  • Some group labels may be inconsistent when Test data is applied to label groups

Notes about Docker on Windows

  • Requires 8GB of memory for docker if running on Windows.

  • Linux it is automatically allocated

Additional Notes

  • ST.BOX now supports DFU mode, without the need for JLINK/v2

Qeexo AutoML 1.11.

New Features:

  • Native App (no need for Chrome extension)

  • Test data management

  • New models: Polynomial Support Vector Machine (POLYSVM), RBF Support Vector Machine (RBFSVM), Gaussian Naive Bayes (GNB), One Class Random Forest (ORF)

  • Event Classification

  • STWIN supports dual microphone

  • PCA plot

  • F1 score

  • Hyperparameter Tuning Optimizer

  • Proxy option for installer

Qeexo AutoML 1.10.3

September 24, 2020

New Features

  • Added new hardware platform that supports M0: Arduino Nano 33 IoT

  • Bronze users can now create custom applications by downloading static library.

Known Issues

  • Specific configurations of multiple sensors when selected together with the microphone, sometimes causes training failures of the CNN model.

  • Selecting data from the 1Hz low-power accelerometer sensor from STWINKT1 hardware sometimes causes training failures.

  • Sometimes, the original data would still exist after re-recording.

  • Class label names cannot contain special characters such as <, >, |, :, “, and \ .

  • Certain USB Type-C hubs cause connection issues in macOS. We recommend using an USB A-to-C adapter rather than a hub.

Qeexo AutoML 1.9.1

September 5, 2020


New Features

  • Added new hardware platform: RA6M3 ML Sensor Module from Renesas. Currently, accelerometer, gyroscope, humidity, light, and temperature sensors are supported.

  • Improved Qeexo AutoML signup process.

  • Added DFU mode support for ST SensorTile.box, which means that the ST-LINK/v2 programmer and adapter are no longer required to flash the SensorTile.box.

Bug Fixes & Improvements

  • Fixed latency calculation failures.

  • Added various UI-related fixes/improvements.

Known Issues

  • Specific configurations of multiple sensors when selected together with the microphone, sometimes causes training failures of the CNN model.

  • Selecting data from the 1Hz low-power accelerometer sensor from STWINKT1 hardware sometimes causes training failures.

  • Sometimes, the original data would still exist after re-recording.

  • Class label names cannot contain special characters such as <, >, |, :, “, and \ .

  • Certain USB Type-C hubs cause connection issues in macOS. We recommend using an USB A-to-C adapter rather than a hub.

Qeexo AutoML 1.8.2

August 10, 2020

New Features

  • Added new hardware platform: STWINKT1 from STMicroelectronics.

  • Added new ML algorithms: one class SVM, linear SVM, RNN, CRNN.

  • Added Live Classification Analysis to finetune trained models.

  • Added quantization options to reduce model size.

Bug Fixes & Improvements

  • Fixed a flashing issue on the Arduino Nano 33 BLE Sense.

  • Fixed a training issue that occurs when only one axis (x, y, or z) is selected for accelerometer and/or gyroscope.

  • Added data-check-related improvements.

  • Added various UI-related fixes/improvements.

Known Issues

  • When different data types (Continuous/Event) are grouped together during training, unexpected results such as training failure may occur. We recommend not mixing different data types within a Group.

  • Specific configurations of multiple sensors when selected together with the microphone, sometimes causes training failures of the CNN model.

  • Class label names cannot contain special characters such as <, >, |, :, “, and \ .

  • Certain USB Type-C hubs cause connection issues in macOS. We recommend using an USB A-to-C adapter rather than a hub.