Qeexo

TRY
BACK TO PRESS

Qeexo AutoML 1.19.0 New Feature Introduction

Cora Zhang 18 January 2023

In this article, we are going to introduce you some of the latest and greatest features and improvements released in Qeexo AutoML 1.19.0.

Feature 1 – ‘FILTER ENVIRONMENT’ feature, DATA page

Image 1 – Where to find the Filter Environment feature
Image 2 – The Filter pop-up window

‘FILTER ENVIRONMENT’ enables you to filter your data collections with different environments settings and sensor configurations settings.

When the user clicks the ‘FILTER ENVIRONMENT’ button, there will be a pop-up window. On the very top area of the window, there is a drop-down menu for users to select to list either all environments or any specific environment. Moving down, there is a two columns table where users will see environment names and sensor configurations settings. Users can click or unclick the checkbox to the left to select any environment(s), then click FILTER button to apply. Once option has been applied, users will be navigated back to DATA page, and only those data collections that match with the selection will be displayed under the DATA page.

Video 1 – Filter Environment

Feature 2 – ‘SELECT ALL’ feature under DATA page

Image 3 – The new data page SELECT ALL feature

‘SELECT ALL’ button enables users to select all data collections listed under DATA page with a single click. This feature is very helpful when users have many data collections to select to train a model. It saves users a lot of time and work. If users have data collections that are sharing different environments or different sensor configurations, users could use ‘FILTER ENVIRONMENT’ button to filter data collections, only keep those that need to be used. Then use the ‘SELECT ALL’ checkbox.

Please note that although users are allowed to click to select all data collections regardless of sensor configuration settings, the START NEW TRAINING button will only become clickable when all selected data collections share the same sensor configuration. If there’s multiple pages of data collections, users will need to SELECT ALL for each page.

Video 2 – Select All

Feature 3 – Upgraded ‘Data Segmentation’ feature under DATA page

Image 4 – The Create New Segment pop-up window

The upgraded Data Segmentation features save users time and work on cropping a data collection into many different segments.

User can click on ‘CREATE NEW SEGMENT’ button to create a segment label, then select a color. After users click the ‘CREATE’ button, users will be able to click the start and end point of a segment on the data plot as many times as needed. If there’s another segment label users would like to create, users could simply repeat the process to crop data with a new segment label.

Image 5 – Data Segmentation
Video 3 – Create new segment

Every time a user creates a segment, the segment will automatically be displayed in the segments list under the data plot.  Users have the option to edit (including name, duration (start and end time)) or delete it.

Image 6 – Segment fine-tuning
Video 4 – Edit segment

To help users match the segments on the plot with the lines under plot, we have added a click and reveal feature to indicate the corresponding segments from the plot when the user clicks the segment’s checkbox.

Image 7 – Selecting a segment automatically highlights it in the graph

Video 5 – Click and reveal segment

Additionally, by clicking the check box to the left of the segment label(s) on the top, it will select all segments. Users could also click the trash button on top right to delete all selected segments at once.

Image 8 – Select all
Video 6 – Select and delete all segments

Feature 4 – ASSISTED SEGMENTATION feature under DATA page

Image 9 – Assisted Segmentation
Image 10 – The Assisted Segmentation pop-up window

Assisted Segmentation provides users with an even quicker way to crop a data collection into segments. Users just need to manually crop a few segments for each label and then use Assisted Segmentation to help detect those data points that share the same information. More specifically, users could follow the steps in Feature 3 – click ‘CREATE NEW SEGMENT’, create a segment label, click the start and end point on data plot to crop a segment. Then users could simply click ‘ASSISTED SEGMENTATION’ button on the top right. There will be a pop-up window for users to select segment label(s) for assisted segmentation. Once users clicks NEXT, Qeexo AutoML will detect the segments for the un-cropped areas and list them on both the data plot and segments list below. Users could either ACCEPT or REJECT the result depending on how satisfied you are with the segmentation.

Image 11 – Assisted Segmentation results

Feature 5 – Advanced option of setting up ML Static Library Memory Constraint

When Pro Tier users select data collections and continue to click START NEW TRAINING button to start the model training, we added a new advanced option to allow users to configure ML Static Library Memory Constraint, more specifically, users could set the target ML Static Library maximum flash and RAM memory size usage as desired. Generally, the model performance will reduce as the model size reduces, but not by a lot. The tradeoff should be worth it.

In the ‘Algorithm Selection’ window, user could click CONFIGURE button to access to the Memory Constraint feature. The toggle on the right side should be enabled before users type in any values in the field on the left side.

Image 12 – Advanced options step 1 – Configure
Image 13 – Advanced options step 2 – Enable with the slider on the right, then enter settings

Feature 6 – Report on size taken by ML Model and Other code separately per ML library built under Model page

Image 14 – Model size reporting

After users trained a model, under performance summary for each ML library (algorithm) built, we added a new feature to report the size occupied by ML model. From here, users could check the flash and RAM size of binary image, ML model, pre-processing + featurization and the total size. The very basic benefit is that it brings more model information for users to use as a reference when selecting the final model. We are sure it also provides more flexibility and possibilities per user case.

Feature 7 – DATA Augmentation Toolkit

Data augmentation is a new feature that we just added for Qeexo AutoML 1.19.0. The major function is to add more data to the training data which could help the model become more robust. This additional data is basically a fictious data which has been created by applying some operations to the original data. We currently support two types of augmentation – scaling and jittering. These two augmentations are enabled for all sensors except environmental sensors.

BACK TO PRESS

TDK to acquire Qeexo to enable complete smart edge platforms

Qeexo 04 January 2023
  • TDK to acquire Qeexo, Co, a leading developer of automated machine-learning (ML) platform that accelerates the development of tinyML models for low power, always-on intelligent platforms
  • TDK aims to further strengthen its ML expertise and simplify ML application development to become a leader in delivering smart edge solutions
  • Acquisition enables TDK to accelerate the transition to Industry 4.0 with smart edge solutions

SAN JOSE, Calif., Jan. 4, 2023 /PRNewswire/ — TDK Corporation (TSE: 6762) (CEO & President: Noboru Saito, hereinafter “TDK”) announced that today TDK has agreed to acquire Qeexo, Co. (CEO: Sang Won Lee, hereinafter “Qeexo”), a U.S.-based venture-backed company spun out of Carnegie Mellon University engaged in the automation of end-to-end machine learning for edge devices. As a result of the acquisition, Qeexo will become a wholly owned subsidiary of TDK, subject to customary closing conditions, including approval of the Committee on Foreign Investment in the US (CFIUS).

Qeexo, based in Mountain View, California, USA, is the first company to automate end-to-end machine learning for edge devices. Qeexo AutoML enables a no-code environment, enabling data collection and training of 18 (and expanding) different machine learning algorithms, including both neural networks and non-neural-networks, to the same dataset, while generating metrics for each (accuracy, memory size, latency), so that users can pick the model that best fits their unique requirements. A cloud-based easy to use solution, it provides an intuitive UI platform system that allows users to collect, annotate, cleanse, and visualize sensor data and automatically build “tinyML” models using different algorithms. Qeexo’s AutoML platform allows customers to leverage sensor data to rapidly build machine learning solutions optimized to have ultra-low latency and power consumption, with an incredibly small memory footprint for highly constrained environments with applications in industrial, IoT, wearables, automotive, mobile, and more.  Through streamlined intuitive process automation, Qeexo’s AutoML enables customers without precious ML resources and greatly accelerates design of Edge AI capabilities for their own specific applications.

“Qeexo brings together a unique combination of expertise in automating machine learning application development and deployment for those without ML expertise, high volume shipment of ML applications and understanding of sensors to accelerate the deployment of smart edge solutions,” stated Jim Tran, CEO, TDK USA Corporation. “Their expertise combined with TDK’s leadership positions in sensors, batteries and other critical components will enable the creation of system level solutions addressing a broad range of applications and industries.”

“Our platform is an outgrowth of our own history of high-volume ML application development and deployment enabling those with domain expertise but not ML expertise to solve real world problems quickly and efficiently,” continued Sang Lee, CEO, Qeexo. “We see our AutoML tool as a natural partner to the smarter sensor systems that TDK is building.”

The following is an outline of the company profile:

  1. Company name: Qeexo, Co.
  2. Location: Headquartered in Mountain View, CA, office in Pittsburgh, PA, USA
  3. Established: September 2012
  4. Management: CEO – Sang Won Lee; CTO – Chris Harrison
  5. Main business operations: Development of automated machine-learning (ML) platform that accelerates the development of tinyML models for the Edge.
  6. Learn more about fundamental machine learning concepts: ­Qeexo AutoML Best Practice Guide – Qeexo, Co.

TDK will be showcasing over 30 different technologies, solutions, and platforms at CES 2023, January 5-8, 2023, at the Las Vegas Convention Center (LVCC) and can be found at Central Hall – #16181. Qeexo will demonstrate their machine learning platform solution within the TDK booth and also showcase their full range of technology solutions at the Qeexo booth #11222, North Hall. 

Glossary

  • AutoML: Automated machine learning is the process of automating the tasks of applying machine learning to real-world problems.
  • tinyML: Tiny machine learning is broadly defined as a fast-growing field of machine learning technologies that is capable of performing on-device sensor data analytics at extremely low power,
  • ML: Machine learning is a field of inquiry devoted to understanding and building methods that ‘learn’, that is, methods that leverage data to improve performance on some set of tasks
  • Smart Edge solutions: Smart Edge solutions refers to the analysis of data and development of solutions at the site where the data is generated.
  • Smart Edge device:  An intelligent edge device is a sophisticated IoT device that performs some degree of data processing within the device itself.

About TDK Corporation
TDK Corporation is a world leader in electronic solutions for the smart society based in Tokyo, Japan. Built on a foundation of material sciences mastery, TDK welcomes societal transformation by resolutely remaining at the forefront of technological evolution and deliberately “Attracting Tomorrow.” It was established in 1935 to commercialize ferrite, a key material in electronic and magnetic products. TDK’s comprehensive, innovation-driven portfolio features passive components such as ceramic, aluminum electrolytic and film capacitors, as well as magnetics, high-frequency, and piezo and protection devices. The product spectrum also includes sensors and sensor systems such as temperature and pressure, magnetic, and MEMS sensors. In addition, TDK provides power supplies and energy devices, magnetic heads and more. These products are marketed under the product brands TDK, EPCOS, InvenSense, Micronas, Tronics and TDK-Lambda. TDK focuses on demanding markets in automotive, industrial and consumer electronics, and information and communication technology. The company has a network of design and manufacturing locations and sales offices in Asia, Europe, and in North and South America. In fiscal 2022, TDK posted total sales of USD 15.6 billion and employed about 117,000 people worldwide.

About Qeexo
Qeexo is the first company to automate end-to-end machine learning for embedded edge devices (Cortex M0-M4 class). Our one-click, fully-automated Qeexo AutoML platform allows customers to leverage sensor data to rapidly build machine learning solutions for highly constrained environments with applications in industrial, IoT, wearables, automotive, mobile, and more. Over 300 million devices worldwide are equipped with AI built on Qeexo AutoML. Delivering high performance, solutions built with Qeexo AutoML are optimized to have ultra-low latency, ultra-low power consumption, and an incredibly small memory footprint.

Images related to this release can be downloaded from the following URL: https://www.tdk.com/en/news_center/press/20230104_01.html

Contacts for regional media

RegionContactPhoneMail
GlobalMr. David A.ALMOSLINOTDK USA Corporation
San Jose, CA
+1 408-501-2278[email protected]  
North AmericaMs. Sarah
MACKENZIE
Publitek
Portland, OR
+1 503-720-3743TDK-[email protected]
JapanMr. Yoichi
OSUGA
TDK Corporation
Tokyo, Japan
+813 6778-1055[email protected]
WorldwideMr. Sang
Won Lee
Qeexo
Mountain View, CA
+1 510 508 0446[email protected] 

SOURCE TDK Corporation