{"id":416408,"date":"2020-02-21T12:17:26","date_gmt":"2017-07-24T13:56:29","guid":{"rendered":"https:\/\/www.microsoft.com\/en-us\/research\/?post_type=msr-project&p=416408"},"modified":"2020-03-11T13:05:49","modified_gmt":"2020-03-11T20:05:49","slug":"resource-efficient-ml-for-the-edge-and-endpoint-iot-devices","status":"publish","type":"msr-project","link":"https:\/\/www.microsoft.com\/en-us\/research\/project\/resource-efficient-ml-for-the-edge-and-endpoint-iot-devices\/","title":{"rendered":"Resource-efficient ML for Edge and Endpoint IoT Devices"},"content":{"rendered":"

Our objective is to develop a library of efficient machine learning algorithms that can run on severely resource-constrained edge and endpoint IoT devices ranging from the Arduino to the Raspberry Pi. Please see our GitHub<\/a> page for code release.<\/p>\n

The Internet of Things (IoT) is poised to revolutionize our world. Billions of microcontrollers and sensors have already been deployed for predictive maintenance, connected cars, precision agriculture, personalized fitness and wearables, smart housing, cities, healthcare, etc. The dominant paradigm in these applications is that the IoT device is dumb\u2014it just senses its environment and transmits the sensor readings to the cloud where all the intelligence resides and the decision making happens.<\/p>\n

We envision an alternative paradigm where\u00a0even tiny, resource-constrained IoT devices can run machine learning algorithms locally<\/i>\u00a0without necessarily connecting to the cloud. This enables a number of critical scenarios, beyond the pale of the traditional paradigm, where it is not desirable to send data to the cloud due to concerns about latency, connectivity, energy, privacy and security.<\/p>\n

We are therefore currently releasing tree and k-nearest neighbour based algorithms, called Bonsai<\/a> and ProtoNN<\/a> respectively, for classification, regression, ranking and other common IoT tasks. Bonsai and ProtoNN can be trained on the cloud, or on a laptop, but can then make predictions locally on the tiniest of microcontrollers without needing cloud connectivity.<\/p>\n

We have deployed Bonsai and ProtoNN on the Arduino Uno (8 bit ATmega328P microcontroller operating at 16 MHz without floating point support, 2 KB RAM and 32 KB read-only flash memory) and found that they can accurately make predictions within a few milliseconds. Detailed experiments and prediction cost-accuracy trade-offs can be found in our ICML 2017 papers.<\/p>\n","protected":false},"excerpt":{"rendered":"

Our objective is to develop a library of efficient machine learning algorithms that can run on severely resource-constrained edge and endpoint IoT devices ranging from the Arduino to the Raspberry Pi.<\/p>\n","protected":false},"featured_media":0,"template":"","meta":{"msr-url-field":"","msr-podcast-episode":"","msrModifiedDate":"","msrModifiedDateEnabled":false,"ep_exclude_from_search":false,"_classifai_error":"","footnotes":""},"research-area":[13556,13547],"msr-locale":[268875],"msr-impact-theme":[],"msr-pillar":[],"class_list":["post-416408","msr-project","type-msr-project","status-publish","hentry","msr-research-area-artificial-intelligence","msr-research-area-systems-and-networking","msr-locale-en_us","msr-archive-status-active"],"msr_project_start":"","related-publications":[393413,393422,394229,486405,546513,550788],"related-downloads":[557907,621708],"related-videos":[],"related-groups":[],"related-events":[],"related-opportunities":[],"related-posts":[],"related-articles":[],"tab-content":[],"slides":[],"related-researchers":[{"type":"user_nicename","display_name":"Ajay Manchepalli","user_id":30885,"people_section":"Section name 1","alias":"ajayma"},{"type":"user_nicename","display_name":"Harsha Simhadri","user_id":36146,"people_section":"Section name 1","alias":"harshasi"},{"type":"user_nicename","display_name":"Manik Varma","user_id":32791,"people_section":"Section name 1","alias":"manik"},{"type":"user_nicename","display_name":"Rahul Sharma","user_id":36308,"people_section":"Section name 1","alias":"rahsha"},{"type":"user_nicename","display_name":"Suresh Parthasarathy","user_id":33762,"people_section":"Section name 1","alias":"supartha"},{"type":"user_nicename","display_name":"Vivek Seshadri","user_id":36323,"people_section":"Section name 1","alias":"visesha"}],"msr_research_lab":[199562],"msr_impact_theme":[],"_links":{"self":[{"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-project\/416408"}],"collection":[{"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-project"}],"about":[{"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/types\/msr-project"}],"version-history":[{"count":15,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-project\/416408\/revisions"}],"predecessor-version":[{"id":644247,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-project\/416408\/revisions\/644247"}],"wp:attachment":[{"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/media?parent=416408"}],"wp:term":[{"taxonomy":"msr-research-area","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/research-area?post=416408"},{"taxonomy":"msr-locale","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-locale?post=416408"},{"taxonomy":"msr-impact-theme","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-impact-theme?post=416408"},{"taxonomy":"msr-pillar","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-pillar?post=416408"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}