693d
693d
693d
693d Posted by Hee Jung, Developer 693d Relations Group Supervisor / Soonson 693d Kwon, Developer Relations Program Supervisor
693d
ML in Motion is a 693d digital occasion to gather and 693d share cool and helpful machine 693d studying (ML) use circumstances that 693d leverage a number of Google 693d ML merchandise. That is the 693d primary run of an ML 693d use case marketing campaign by 693d the ML Developer Packages crew.
693d
693d
Allow us to announce the 693d winners proper now, proper right 693d here. They’ve showcased sensible makes 693d use of of ML, and 693d the way ML was tailored 693d to actual life conditions. We 693d hope these tasks can spark 693d new utilized ML undertaking concepts 693d and supply alternatives for ML 693d group leaders to debate ML 693d use circumstances.
693d
693d
4 Winners of “ML in 693d Motion” are:
693d
693d Detecting Meals High quality with 693d Raspberry Pi and TensorFlow
693d
693d
693d By George Soloupis, ML Google 693d Developer Skilled (Greece)
693d
693d
This undertaking helps folks with 693d odor impairment by figuring out 693d meals degradation. The thought got 693d here immediately when a pal 693d revealed that he has no 693d sense of odor attributable to 693d a motorcycle crash. Even with 693d experiences attending quite a lot 693d of IT conferences, this concern 693d was unaddressed and the facility 693d of machine studying is one 693d thing we might depend on. 693d Therefore the objective. It’s to 693d create a prototype that’s inexpensive, 693d correct and usable by folks 693d with minimal data of computer 693d systems.
693d
693d
The fundamental setting of the 693d meals high quality detection is 693d that this. Raspberry Pi collects 693d knowledge from air sensors over 693d time through the meals degradation 693d course of. This single board 693d pc was very helpful! With 693d the GUI, it’s straightforward to 693d execute Python scripts and see 693d the outcomes on display screen. 693d Eight sensors collected knowledge of 693d the chemical components comparable to 693d NH3, H2s, O3, CO, and 693d CH4. After working the prototype 693d for in the future, classes 693d had been set following the 693d outcomes. The primary hours of 693d the meals out of the 693d fridge as “good” and the 693d remainder as “unhealthy”. Then the 693d dataset was evaluated with the 693d assistance of TensorFlow and the 693d inference was executed with TensorFlow 693d Lite.
693d
693d
Since there have been no 693d open supply prototypes on the 693d market with related targets, it 693d was an entire journey. Sensors 693d on PCBs and standalone sensors 693d had been used to get 693d the perfect combination of accuracy, 693d stability and sensitivity. A logic 693d degree converter has been used 693d to reduce the usage of 693d resistors, and capacitors have been 693d positioned for stability. And the 693d consequence, a compact prototype! The 693d Raspberry Pi might connect instantly 693d on with slots for eight 693d sensors. It’s developed in such 693d a method that sensors could 693d be changed at any time. 693d Customers can experiment with totally 693d different sensors. And the inference 693d time values are despatched via 693d the bluetooth to a cell 693d system. As an finish consequence 693d a consumer with no superior 693d technical data will have the 693d ability to see meals high 693d quality on an app constructed 693d on Android (Kotlin).
693d
693d
Reference: 693d Github 693d , 693d extra to learn 693d
693d
693d
* This undertaking is supported 693d by 693d Google Affect Fund 693d .
693d
693d Election Watch: Making use of 693d ML in Analyzing Elections Discourse 693d and Citizen Participation in Nigeria 693d
693d
693d
693d By Victor Dibia, ML Google 693d Developer Skilled (USA) 693d
693d
693d
693d
This undertaking explores the usage 693d of GCP instruments in ingesting, 693d storing and analyzing knowledge on 693d citizen participation and election discourse 693d in Nigeria. It started on 693d the premise that the proliferation 693d of social media interactions offers 693d an attention-grabbing lens to check 693d human conduct, and ask necessary 693d questions on election discourse in 693d Nigeria in addition to interrogate 693d social/demographic questions.
693d
693d
It’s based mostly on knowledge 693d collected from twitter between September 693d 2018 to March 2019 (tweets 693d geotagged to Nigeria and tweets 693d containing election associated key phrases). 693d Total, the info set incorporates 693d 25.2 million tweets and retweets, 693d 12.6 million authentic tweets, 8.6 693d million geotagged tweets and 693d three.6 million tweets labeled (utilizing 693d an ML mannequin) as political. 693d
693d
693d
By analyzing election discourse, we 693d are able to be taught 693d a couple of necessary issues 693d together with – points that 693d drive election discourse, how social 693d media was utilized by candidates, 693d and the way participation was 693d distributed throughout geographic areas within 693d the nation. Lastly, in 693d a rustic like Nigeria the 693d place up to date demographics 693d knowledge is missing (e.g., on 693d group buildings, wealth distribution and 693d so forth), this undertaking reveals 693d how social media can be 693d utilized as a surrogate to 693d deduce relative statistics (e.g., existence 693d of diaspora communities based mostly 693d on election dialogue and wealth 693d distribution based mostly on system 693d kind utilization throughout the nation).
693d
693d
Knowledge for the undertaking was 693d collected utilizing python scripts that 693d wrote tweets from the Twitter 693d streaming api (matching sure standards) 693d to BigQuery. BigQuery queries had 693d been then used to generate 693d mixture datasets used for visualizations/evaluation 693d and coaching machine studying fashions 693d (political textual content classification fashions 693d to label political textual content 693d and multi class 693d classification fashions to label normal 693d discourse). The fashions had been 693d constructed utilizing Tensorflow 2.0 693d and educated on 693d Colab notebooks 693d powered by 693d GCP GPU compute VMs 693d .
693d
693d
References: 693d Election Watch web site 693d , ML fashions descriptions 693d one 693d , 693d two 693d
693d
693d Bioacoustic Sound Detector (To establish 693d chook calls in soundscapes)
693d
693d
693d By Usha Rengaraju, TFUG Organizer 693d (India)
693d
693d
“Visionary Perspective Plan (2020-2030) for 693d the conservation of avian range, 693d their ecosystems, habitats and landscapes 693d within the nation” proposed by 693d the Indian authorities to assist 693d in the conservation of birds 693d and their habitats impressed me 693d to take up this undertaking.
693d
693d
Extinction of chook species is 693d an growing international concern because 693d it has a huge effect 693d on meals chains. Bioacoustic monitoring 693d can present a passive, low 693d labor, and cost-effective technique for 693d finding out endangered chook populations. 693d Current advances in machine studying 693d have made it attainable to 693d mechanically establish chook songs for 693d frequent species with ample coaching 693d knowledge. This innovation makes it 693d simpler for researchers and conservation 693d practitioners to precisely survey inhabitants 693d tendencies they usually’ll have the 693d ability to usually and extra 693d successfully consider threats and regulate 693d their conservation actions.
693d
693d
This undertaking is an implementation 693d of a Bioacoustic monitor utilizing 693d Masked Autoencoders in TensorFlow and 693d Cloud TPUs. The undertaking might 693d be introduced as a browser 693d based mostly utility utilizing Flask. 693d The deep studying prototype can 693d course of steady audio knowledge 693d after which acoustically acknowledge the 693d species.
693d
693d
The objective of the undertaking 693d once I began was to 693d construct a primary prototype for 693d monitoring of uncommon chook species 693d in India. In future I 693d wish to develop the undertaking 693d to observe different endangered species 693d as effectively.
693d
693d
References: 693d Kaggle Pocket book 693d , 693d Colab Pocket book 693d , 693d Github 693d , 693d the dataset 693d and 693d extra to learn 693d
693d
693d Persona Labs’ Digital Personas
693d
693d
693d By Martin Andrews and Sam 693d Witteveen, ML Google Developer Specialists 693d (Singapore) 693d
693d
693d During the last 3 years, 693d Pink Dragon AI (an organization 693d co-founded by Martin and Sam) 693d has been growing real-time digital 693d “Personas”. The important thing concept 693d is to allow customers to 693d work together with life-like Personas 693d in a format just like 693d a Zoom name : Talking 693d to them and seeing them 693d reply in actual time, simply 693d as a human would. Naturally, 693d every Persona could be tailor-made 693d to duties required (by adjusting 693d the looks, voice, and ‘motivation’ 693d of the dialog system behind 693d the scenes and their corresponding 693d backend APIs).693d
693d
The parts required to make 693d the Personas work successfully embrace 693d dynamic face fashions, expression era 693d fashions, Textual content-to-Speech (TTS), dialog 693d backend(s) and Speech Recognition (ASR). 693d A lot of this was 693d constructed on GCP, with GPU 693d VMs working the (many) Deep 693d Studying fashions and mixing the 693d outputs into dynamic WebRTC video 693d that streams to customers through 693d a browser front-end.
693d
693d
A lot of the earlier 693d years’ work focussed on making 693d the Personas’ faces behave in 693d a life-like method, whereas ensuring 693d that the general latency (i.e. 693d the time between the Persona 693d listening to the consumer asking 693d a query, to their lips 693d beginning the response) is stored 693d low, and the rendering 693d of particular person photographs matches 693d the 25 frames-per-second video charge 693d required. As you may think, 693d there have been many Deep 693d Studying modeling challenges, coupled with 693d laborious engineering points to beat.
693d
693d
By way of backend applied 693d sciences, Google Cloud GPUs had 693d been used to coach the 693d Deep Studying fashions (constructed utilizing 693d TensorFlow/TFLite, PyTorch/ONNX & extra not 693d too long ago JAX/Flax), and 693d the real-time serving is finished 693d by Nvidia T4 GPU-enabled VMs, 693d launched as required. Google ASR 693d is at present used as 693d a streaming backend for speech 693d recognition, and Google’s WaveNet TTS 693d is used when multilingual TTS 693d is required. The system 693d additionally makes use of Google’s 693d serverless stack with CloudRun and 693d Cloud Capabilities being utilized in 693d a few of the dialog 693d backends.
693d
693d
Go to the Persona’s web 693d site (linked beneath) and you 693d may see movies that exhibit 693d a number of features : 693d What the Personas appear like; 693d their Multilingual functionality; potential purposes; 693d and so forth. 693d Nonetheless, the movies can’t actually 693d exhibit what the interactivity ‘appears 693d like’. For that, it’s 693d finest to get a dwell 693d demo from Sam and Martin 693d – and see what real-time 693d Deep Studying mannequin era seems 693d like!
693d
693d
Reference: 693d The Persona Labs web site 693d
693d
693d