Bonsai Algorithm Enables Machine Learning on Arduino with a 2KB RAM Footprint

Bonsai Machine Learning

Machine learning used to be executed in the cloud, then the inference part  moved to the edge, and we’ve even seen micro-controllers able to do image recognition with GAP8 RISC-V micro-controller. But I’ve recently come across a white paper entitled “Resource-efficient Machine Learning in 2 KB RAM for the Internet of Things” that shows how it’s possible to perform such tasks with very little resources. Here’s the abstract: This paper develops a novel tree-based algorithm, called Bonsai, for efficient prediction on IoT devices – such as those based on the Arduino Uno board having an 8 bit ATmega328P microcontroller operating at 16 MHz with no native floating point support, 2 KB RAM and 32 KB read-only flash. Bonsai maintains prediction accuracy while minimizing model size and prediction costs by: (a) developing a tree model which learns a single, shallow, sparse tree with powerful nodes; (b) sparsely projecting all data into a low-dimensional space in which the tree is learnt; and …

Support CNX Software – Donate via PayPal or become a Patron on Patreon

Microsoft Surface Go Tablet Based on Intel Pentium Gold 4415Y Processor to Sell for $399 and Up

Microsoft Surface Go

Microsoft Surface devices are powerful but usually expensive ($2,000+) tablets and laptops, but the company has now launched an entry-level model called Surface Go, powered by an Intel Kaby Lake Pentium processor with 4 to 8GB RAM, 64 to 256 GB storage, and a 10″ high resolution display, and selling for $399 and up. Microsoft Surface Go specifications: SoC – Intel Pentium Gold 4415Y dual core/four thread Kaby Lake-Y processor @ up to 1.6 GHz (base frequency), with  24EU Intel HD Graphics 615 @ 300 / 850 MHz; 6W TDP System Memory – 4 or 8GB LPDDR3-1866 Storage –  64GB eMMC flash, 128GB or 256GB SSD, micro SD card slot Display – 10″ 1800×1200 PixelSense display; 3:2 aspect ratio; built-in kickstand with full friction hinge that extends to 165 degrees Cameras – 8MP Rear camera, 5MP Front-facing camera with Windows Hello facial recognition Audio – 3.5mm headphone jack Connectivity – 802.11.ac Wi-Fi + Bluetooth 4.1; LTE model(s) to be announced …

Support CNX Software – Donate via PayPal or become a Patron on Patreon

Visual Studio 2017 with an Embedded Linux Arm Device

This is a non-sponsored guest post written by Marc Goodner: Principal Program Manager, Microsoft, and Jeremias Cordoba: Innovation Engineer, Toradex. Today many embedded devices run some flavor of Linux as their primary operating system. This poses a challenge to developers who run Windows on their development machine. This article explains a new way to use the latest Visual Studio for C++ development on an embedded Arm Devices from a Windows Host PC using containers for the build environment. The device we are deploying to is from the Toradex Colibri Family of System on Modules using the NXP i.MX 6ULL SoC, which features an Arm Cortex A-7. As a demo project we will connect a Bluetooth Sensor with the Toradex Colibri Module. Please note that Visual Studio support for this case is in an early state, you will see improvements from Microsoft and Toradex in the coming months. Prerequisites Colibri i.MX 6ULL with Wi-Fi/BT and an Aster Carrier Board TI SensorTag (Bluetooth low energy) …

Support CNX Software – Donate via PayPal or become a Patron on Patreon

HP t430 Thin client Works with a Single USB Cable

Gemini Lake mini PCs are starting to sell and ship, and from the look of it, HP t430 appears to be yet another Gemini Lake mini PC based on an Intel Celeron N4000 processor. But the device is actually a thin client that has been designed to work with a single USB type C cable for power, video and audio. HP t430 specifications: SoC – Intel Celeron N4000 dual core processor @ 1.1 GHz / 2.4 GHz with Intel UHD Graphics 600 System Memory – Up to 4GB DDR4 RAM Storage – 16 GB to 32 GB Flash memory Display – HDMI, DisplayPort 1.2, USB type C DisplayPort Alt-mode Audio – Headphone/microphone mini-jack Networking – Gigabit Ethernet, and WLAN USB – 3x USB type A ports, 1x USB type C (DisplayPort 1.2, USB 3.1, charging) Misc – Power button, Kensington lock Power Supply – 19V DC, 45 W, worldwide auto-sensing, 100-240 VAC, 50-60 Hz, energy saving automatic power-down, surge tolerant external …

Support CNX Software – Donate via PayPal or become a Patron on Patreon

MT3620 Development Board for Azure Sphere is Up for Pre-order for $84.90

Microsoft unveiled Azure Sphere solutions last month in order to provide a complete secure IoT solution from the hardware to the cloud with three main components: a secure MCU, Linux based Azure Sphere OS, and Azure Sphere Security (Cloud) Service. The company already stated Mediatek MT3620 would be the first Azure Sphere compliant MCU,  and today Seeed Studio has announced MT3620 development board for Azure Sphere is up for pre-order for $84.90 with a $5 deposit, and shipping scheduled for the end of September. Preliminary specifications of MT3620 development board for Azure Sphere: WiSoC – Mediatek MT3620AN single core Arm Cortex-A7 processor @ 500 MHz, dual core Arm Cortex-M4F real-time core, Pluton security sub-system, and WiFi. System Memory – Over 5MB of embedded RAM, split among the various cores (4MB for Arm Cortex A7 core) Storage – 2x 8MB dual channel quad SPI (16MB in total) Connectivity –  Dual band 802.11 a/b/g/n WiFi 1T1R with main and aux PCB antennas, …

Support CNX Software – Donate via PayPal or become a Patron on Patreon

Microsoft Introduces Project Kinect for Azure

Kinect started as as motion sensing input device for Xbox 360 & One game consoles with an integrated webcam, infrared projector, and a microphone array, which enabled it to detect depth, motion, and voice from gamers. But it also become popular with maker projects due to price and capabilities, and was even found to be a worthy microphone array for voice commands. The third generation of Kinect went into Hololense mixed reality headset, and Microsoft has just unveiled the fourth generation Kinect with Project Kinect for Azure. AFAIK, the company did not provide the full technical details about the new Kinect, but we know it comes with an RGB camera, a 360-degree mic array, an accelerometer, and a time-of-flight (ToF) depth-sensor, and the company did offer some details about the latter : Number of pixels: 1024×1024 resolution Highest Figure of Merit (highest modulation frequency and modulation contrast resulting in low power consumption with overall system power of 225-950mw) Automatic per …

Support CNX Software – Donate via PayPal or become a Patron on Patreon

Microsoft Introduces Azure Sphere OS Linux Operating System for Certified Microcontrollers (MediaTek MT3620 for now)

Last year, Microsoft announced they were working with Mediatek on Project Sopris Secure WiFi MCU, based on Mediatek MT7687 MIPS microcontroller with 7 properties enhancing security (Hardware-based Root of Trust, failure reporting etc…). The company has now unveiled a complete secure IoT solution with Azure Sphere comprised of three main components: Azure Sphere certified secured MCUs which combine real-time and application cores with built-in Microsoft security technology and connectivity. The first compatible MCU will be Mediatek MT3620, which differs from MIPS based Project Sopris, as it features one Cortex A7 and two Cortex M4 cores as well as WiFi connectivity. Azure Sphere OS that provides  four layers on top of the hardware: security  monitor, custom Linux kernel, on-chip connectivity services to Azure Sphere Security Service, and app containers for computer (A7 core) or/and real-time I/Os (M4 cores). Azure Sphere Security Service, a secure cloud service for Azure Sphere device that brokers trust for device-to-device and device-to-cloud communication through certificate-based authentication. …

Support CNX Software – Donate via PayPal or become a Patron on Patreon

Qualcomm QCS603 / QCS605 “IoT” SoCs are Designed for AI and Computer Vision Applications

Qualcommn has unveiled the “Qualcomm Vision Intelligence Platform”, which aims at IoT devices with camera leveraging artificial intelligence and computer vision. The first SoCs part of the platform are QCS605 and QCS603 manufactured with a 10nm process, and equipped with an “advanced image signal processor” and the Qualcomm Artificial Intelligence (AI) Engine, as well Arm CPU cluster, Adreno GPU, and Hexagon DSP. QCS603 & QCS605 specifications: CPU QCS603 – 2x 1.6GHz Qualcomm Kryo 300 Gold cores, 6x 1.7GHz Qualcomm Kryo 300 Silver cores QCS605 – 2x 2.5GHz Qualcomm Kryo 300 Gold cores, 6x 1.7GHz Qualcomm Kryo 300 Silver cores Qualcomm Artificial Intelligence Engine DSP Qualcomm Hexagon 685 Vector Processor 2x Qualcomm Hexagon Vector eXtensions (HVX) GPU – Qualcomm Adreno 615 with OpenGL ES 3.2, Vulkan, OpenCL support Neural Processing – Qualcomm Snapdragon Neural Processing Engine programming interface with support for Tensorflow, Caffe/Caffe2, ONNE, Android NN; 2.1 TOPS @ 1w Memory I/F – 16-bit LPDDR4x @ up to 1866MHz Connectivity WiFi …

Support CNX Software – Donate via PayPal or become a Patron on Patreon