Applied artificial intelligence for startups gives an idea about what is really AI beyond the buzzword. What are the main steps and questions you have to ask yourself when building a machine learning product. Throughout the presentation, we illustrate what we are saying through the creation of a fake startup called "IsItSafe.ai".
This workshop has been created for the COOPERATHON that organizes the largest open innovation challenges in Canada.
The content is inspired by our own experience working in the AI field as well as working and collaborating with startups and entrepreneurs throughout Canada, the US and France.
Some sections of this course are inspired from the awesome Full Stack Deep Learning (FSDL) course from UC Berkeley. For these sections, we give full credits to the amazing FSDL community.
Without a recall strategy good companies can go out of business when products must be recalled. This presentation offers advice on how to create a meaningful recall strategy.
DataRobot 머신러닝 자동화 플랫폼은 전 세계 Top Data Scientist 들의 지식, 경험 및 모범 사례를 바탕으로 최고 수준의 자동화와 사용 편리성을 확보한 가장 진보된 머신러닝 자동화 솔루션 입니다. DataRobot을 통해 비즈니스 관계자, 분석가 및 데이터 과학자 등 기술 수준과 관계 없이 모든 사용자가 기존 모델링 기법에 비해 아주 빠르게, 매우 정확한 예측 모델을 수립하고 구축, 관리할 수 있습니다.
Big data in foods & IBM Chef Watson agrofood parkAnders Quitzau
Presentation given @Agrofood Park on May 28, 2015. The presentation defines big data in relation to agriculture and foods, challenges and opportunities. And as a pragmatic, yet sophisticated and hi-tech example, describes how IBM Chef Watson applies big data when creating innovative new recipes.
Without a recall strategy good companies can go out of business when products must be recalled. This presentation offers advice on how to create a meaningful recall strategy.
DataRobot 머신러닝 자동화 플랫폼은 전 세계 Top Data Scientist 들의 지식, 경험 및 모범 사례를 바탕으로 최고 수준의 자동화와 사용 편리성을 확보한 가장 진보된 머신러닝 자동화 솔루션 입니다. DataRobot을 통해 비즈니스 관계자, 분석가 및 데이터 과학자 등 기술 수준과 관계 없이 모든 사용자가 기존 모델링 기법에 비해 아주 빠르게, 매우 정확한 예측 모델을 수립하고 구축, 관리할 수 있습니다.
Big data in foods & IBM Chef Watson agrofood parkAnders Quitzau
Presentation given @Agrofood Park on May 28, 2015. The presentation defines big data in relation to agriculture and foods, challenges and opportunities. And as a pragmatic, yet sophisticated and hi-tech example, describes how IBM Chef Watson applies big data when creating innovative new recipes.
DT is a healthcare technology provider focusing on the Pharmaceutical industry with over 8 years of experience. We already have the likes of GSK,Roche, Sanofi and many other pharmaceuticals as our repeat customers.
Our eDetailing suite has been our cornerstone product, which is already up and running in many departments with over 2,000 users. We have worked on other mobile applications within the pharmaceutical industry that includes dosage calculators, knowledge base programs and many more.
Enigma is a Global Digital Transformation Agency
We build forward thinking bespoke digital business transformation models & solutions where the main competitors can’t deliver.
It’s the synergy of real time strategy and analytics combined to offline data sets that delivers an immersive experience for your needs.In today’s market where our customers have more choices than ever and competition fierce on their heels, they need to know what’s relevant, what interests and what inspires them as best as possible.
Your data might not tell you everything you need to know, but it’s the best place to start.
The rest, that’s all in the approach you take to execute a strategy or communication and how enticing your engagement is. It’s the interpretation of insight that makes
the real difference.
AI-SDV 2020 Conference and exhibition is planned to be held at the Hotel Aston la Scala at the same time physically and virtually. Speakers, attendees and exhibitors are invited to join physically or virtually. We are currently developing a range of digital formats that respond precisely to the speakers, attendees and exhibitors needs, enabling individuals and exhibitors from AI industry and related sectors worldwide to participate in the AI-SDV 2020 in Nice.
Due to rapidly changing developments, it is essential that the overall situation be reviewed on an ongoing basis and any measures taken are updated to reflect the evolving legal requirements.
The Artificial Intelligence Conference on Search, Data and Text Mining, Analytics and Visualization.
The 2020 AI-SDV Conference in Nice, 5. - 6. October 2020
2020 Analytics and Visualization Meeting (Presented by VantagePoint), 6. and 7. October 2020
AI-SDV 2020 is the place to be for everyone involved in advanced search and data applications, text mining and visualization technology. Individuals and companies that are shaping the future of this exciting space that surrounds us and impacts us all will present their latest research findings, tech developments and vision for the future.
It’s a full-on two days of learning, networking and exploring technologies and concepts that are now and will continue to change way we as individuals and organisations work, rest and play. The focus is Artificial Intelligence (AI), Digitization 2.0 (about making companies, processes and people ready for AI), Deep Learning and other topics identified by our community, who are specialists working in scientific and technical information.
The event features around 22 speakers over two days, plus an exhibition to complement the conference programme.
In 2020, based on attendee feedback, we are combining all the concepts the ICIC and II-SDV-Meetings in this one event in Nice.
As past attendees tell us, it’s a fabulous location and, beyond the formal meetings taking place in the conference and exhibition space, we make the most of this gorgeous location for informal evening receptions under the stars and networking dinners.
The size of the group makes it perfect for individuals from all sides of the information and tech space to meet and spend time together formally and also socially.
According to our exhibitors and attendees, the size of the event, the beautiful location and mix of people is why they return year after year. It’s where they learn, stay on top of developments, meet peers, find new collaboration opportunities and do business.
That’s why they return – and we hope you will join us too in 2020 and contribute to our programme as an attendee, speaker or exhibitor – and we look forward to welcoming you.
AI technology is being implemented virtually by every industry in some form or the other, which along with business innovations has also lead to AI ethical issues popping up. In this article, we talk about the ethical challenges in AI, and how we at Repustate, manage these issues while developing our products and solutions.
Shift AI 2020: Building AI-first Products - Ehsan Yousefzadeh (AIG Investments)Shift Conference
Shift AI was a success, connecting hundreds of professionals that were eager to propel the progress of AI and discuss the newest technologies in data mining, machine learning and neural networks. More at https://ai.shiftconf.co/.
Talk description:
With rapid advancements in AI research, new breakthroughs lead to new product opportunities. In this presentation, I will discuss the AI product development process, along with the challenges and rewards that come with it. I’ll also discuss the various degrees of AI products, the teams needed to build them, and what it takes to be a great AI-first product manager.
Building an AI Startup: Realities & TacticsMatt Turck
AI is all the rage in tech circles, and the press is awash in tales of AI entrepreneurs striking it rich after being acquired by one of the giants. As always, the realities of building a startup are different, and the path to success requires not just technical prowess but also thoughtful market positioning and business excellence.
In a talk of interest to anyone building or implementing an AI product, Matt Turck and Peter Brodsky leverage hundreds of conversations with AI (and big data) founders and hard-learned lessons building companies from the ground up to highlight successful strategies and tactics.
Topics include:
Successful data acquisition strategies
Data network effects
Competing with the giants
A pragmatic approach to building an AI team
Why social engineering is just as important to success as groundbreaking AI technology
Barcelona Digital Festival 28th Nov 2019 - Data Analytics in eSports. UbeatCa...CIO Edge
Taken from our BCN Digital Festival last week, for info on attending, speaking or sponsoring our next event on the 29/30th April 2020 email enquiry@digitalenterprisefest.com
Data Analytics in eSports. UbeatCase Study
Building AI & Automate services need a solid base on Data Management, but the current environment is volatile, uncertain, complex and ambiguous so you never know what data will be important in the following months.
The Data Management Platform in an extremely dynamic market like eSports where everything is currently being created, in reinvention and is to be validated is even more challenging.
UBEAT is the leading streaming platform of eSports related content. Created in November 2018, it still hasn’t 12 months of existence but a lot of learnings in its rear mirror and a lot of future to come in its high beam. Especially regarding Data Management.
To apply to speak or sponsor our 2020 events goto www.digitalenterprisefest.com
Take the Big Data Challenge - Take Advantage of ALL of Your Data 16 Sept 2014pietvz
A customer service call can transform internal processes. Information in Tweets and reviews can lead to better products. Structured and unstructured data brought together can reveal patterns and relationships that unlock powerful business opportunities. We will discuss real-world use cases and best practices for building the infrastructure you need to power Big Data analytics solutions. From the latest in Hadoop innovation, cognitive computing, and cloud-based analytical web services, you will learn how organizations large and small can harness the power of unstructured human information to create, deploy, and deliver the next generation of analytics applications.
Powered by HP IDOL, HP Autonomy delivers intelligent applications that allow your organization to understand the concepts and context of all information in real time, mitigating risk and identifying opportunity. Join us at this session to learn how HP Autonomy can unlock the value of your company’s structured and unstructured data for better insight and greater competitive advantage. HP IDOL, the OS for human information, enables you to index, manage, and process all your data, both structured and unstructured. Learn how HP IDOL delivers unprecedented insights into optimized architecture, scalability, performance, mapped security, and connectivity. Find out more about IDOLOnDemand.com and how you can leverage this revolutionary technology in your own organization.
Bio IT World 2019 - AI For Healthcare - Simon Taylor, LucidworksLucidworks
Presentation from Bio IT World, Boston | April 16-18, 2019
Track: AI for Healthcare: Practical Application of AI in Clinical Healthcare
Session Title: To AI or Not to AI, That Is the Question
Speaker: Simon Taylor, Lucidworks
Designing a Successful Governed Citizen Data Science StrategyDATAVERSITY
To compete in today’s digital economy, enterprises require new ways to expand AI across their entire organization. Nearly all firms want to do more with data science, but they don't know where to begin or how to properly empower citizen data scientists to avoid common AI gone wrong accidents.
In this session, we will discuss how to approach your journey into citizen data science with existing analytics talent. Proven best practices and lessons learned from successful early adopters of augmented data science will be shared. We will walk through example initiative roadmaps, recommended staffing, upskilling, mentoring and ongoing governance.
Microlent System is one of the leading software development company in India.
Microlent System has 9+ Years experience in IT Industry.
we have Industry experts for your business consult IT services solutions.
Services :
Mobile App Development
Web development
IT Services
ERP Development
QA and testing
Software Development Service
Read More :
https://microlent.com/productportfolio.pdf
With the Industry 4.0 technologies, enterprises build digital models of the ongoing processes in the companies, reveal the causes of the current situation, predict future scenarios, and plan changes to be adopted.
91% of industrial enterprises have been already investing in digital factories.
Industrial Internet, Should I be Interested?ionSign Oy
The Industrial Internet of Things - or the Smart Connected Products - are emerging and transforming both competition and companies. What's it all about and why should I care?
SAP Hybris solutions are all about providing a connected front office. But the customer experience can easily get damaged if the data from your business partners or end customers is not secure. With the new EU General Data Protection Regulation (GDPR) coming into effect in May 2018, the need to protect your customers’ data is essential for your business. Learn how to reduce cost by integrating security into your implementation process to be ahead of the curve for future cyberattacks.
Salma Karina Hayat is Conscious Digital Transformation Leader at Kudos | Empowering SMEs via CRM & Digital Automation | Award-Winning Entrepreneur & Philanthropist | Education & Homelessness Advocate
When listening about building new Ventures, Marketplaces ideas are something very frequent. On this session we will discuss reasons why you should stay away from it :P , by sharing real stories and misconceptions around them. If you still insist to go for it however, you will at least get an idea of the important and critical strategies to optimize for success like Product, Business Development & Marketing, Operations :)
Reflect Festival Limassol May 2024.
Michael Economou is an Entrepreneur, with Business & Technology foundations and a passion for Innovation. He is working with his team to launch a new venture – Exyde, an AI powered booking platform for Activities & Experiences, aspiring to revolutionize the way we travel and experience the world. Michael has extensive entrepreneurial experience as the co-founder of Ideas2life, AtYourService as well as Foody, an online delivery platform and one of the most prominent ventures in Cyprus’ digital landscape, acquired by Delivery Hero group in 2019. This journey & experience marks a vast expertise in building and scaling marketplaces, enhancing everyday life through technology and making meaningful impact on local communities, which is what Michael and his team are pursuing doing once more with Exyde www.goExyde.com
DT is a healthcare technology provider focusing on the Pharmaceutical industry with over 8 years of experience. We already have the likes of GSK,Roche, Sanofi and many other pharmaceuticals as our repeat customers.
Our eDetailing suite has been our cornerstone product, which is already up and running in many departments with over 2,000 users. We have worked on other mobile applications within the pharmaceutical industry that includes dosage calculators, knowledge base programs and many more.
Enigma is a Global Digital Transformation Agency
We build forward thinking bespoke digital business transformation models & solutions where the main competitors can’t deliver.
It’s the synergy of real time strategy and analytics combined to offline data sets that delivers an immersive experience for your needs.In today’s market where our customers have more choices than ever and competition fierce on their heels, they need to know what’s relevant, what interests and what inspires them as best as possible.
Your data might not tell you everything you need to know, but it’s the best place to start.
The rest, that’s all in the approach you take to execute a strategy or communication and how enticing your engagement is. It’s the interpretation of insight that makes
the real difference.
AI-SDV 2020 Conference and exhibition is planned to be held at the Hotel Aston la Scala at the same time physically and virtually. Speakers, attendees and exhibitors are invited to join physically or virtually. We are currently developing a range of digital formats that respond precisely to the speakers, attendees and exhibitors needs, enabling individuals and exhibitors from AI industry and related sectors worldwide to participate in the AI-SDV 2020 in Nice.
Due to rapidly changing developments, it is essential that the overall situation be reviewed on an ongoing basis and any measures taken are updated to reflect the evolving legal requirements.
The Artificial Intelligence Conference on Search, Data and Text Mining, Analytics and Visualization.
The 2020 AI-SDV Conference in Nice, 5. - 6. October 2020
2020 Analytics and Visualization Meeting (Presented by VantagePoint), 6. and 7. October 2020
AI-SDV 2020 is the place to be for everyone involved in advanced search and data applications, text mining and visualization technology. Individuals and companies that are shaping the future of this exciting space that surrounds us and impacts us all will present their latest research findings, tech developments and vision for the future.
It’s a full-on two days of learning, networking and exploring technologies and concepts that are now and will continue to change way we as individuals and organisations work, rest and play. The focus is Artificial Intelligence (AI), Digitization 2.0 (about making companies, processes and people ready for AI), Deep Learning and other topics identified by our community, who are specialists working in scientific and technical information.
The event features around 22 speakers over two days, plus an exhibition to complement the conference programme.
In 2020, based on attendee feedback, we are combining all the concepts the ICIC and II-SDV-Meetings in this one event in Nice.
As past attendees tell us, it’s a fabulous location and, beyond the formal meetings taking place in the conference and exhibition space, we make the most of this gorgeous location for informal evening receptions under the stars and networking dinners.
The size of the group makes it perfect for individuals from all sides of the information and tech space to meet and spend time together formally and also socially.
According to our exhibitors and attendees, the size of the event, the beautiful location and mix of people is why they return year after year. It’s where they learn, stay on top of developments, meet peers, find new collaboration opportunities and do business.
That’s why they return – and we hope you will join us too in 2020 and contribute to our programme as an attendee, speaker or exhibitor – and we look forward to welcoming you.
AI technology is being implemented virtually by every industry in some form or the other, which along with business innovations has also lead to AI ethical issues popping up. In this article, we talk about the ethical challenges in AI, and how we at Repustate, manage these issues while developing our products and solutions.
Shift AI 2020: Building AI-first Products - Ehsan Yousefzadeh (AIG Investments)Shift Conference
Shift AI was a success, connecting hundreds of professionals that were eager to propel the progress of AI and discuss the newest technologies in data mining, machine learning and neural networks. More at https://ai.shiftconf.co/.
Talk description:
With rapid advancements in AI research, new breakthroughs lead to new product opportunities. In this presentation, I will discuss the AI product development process, along with the challenges and rewards that come with it. I’ll also discuss the various degrees of AI products, the teams needed to build them, and what it takes to be a great AI-first product manager.
Building an AI Startup: Realities & TacticsMatt Turck
AI is all the rage in tech circles, and the press is awash in tales of AI entrepreneurs striking it rich after being acquired by one of the giants. As always, the realities of building a startup are different, and the path to success requires not just technical prowess but also thoughtful market positioning and business excellence.
In a talk of interest to anyone building or implementing an AI product, Matt Turck and Peter Brodsky leverage hundreds of conversations with AI (and big data) founders and hard-learned lessons building companies from the ground up to highlight successful strategies and tactics.
Topics include:
Successful data acquisition strategies
Data network effects
Competing with the giants
A pragmatic approach to building an AI team
Why social engineering is just as important to success as groundbreaking AI technology
Barcelona Digital Festival 28th Nov 2019 - Data Analytics in eSports. UbeatCa...CIO Edge
Taken from our BCN Digital Festival last week, for info on attending, speaking or sponsoring our next event on the 29/30th April 2020 email enquiry@digitalenterprisefest.com
Data Analytics in eSports. UbeatCase Study
Building AI & Automate services need a solid base on Data Management, but the current environment is volatile, uncertain, complex and ambiguous so you never know what data will be important in the following months.
The Data Management Platform in an extremely dynamic market like eSports where everything is currently being created, in reinvention and is to be validated is even more challenging.
UBEAT is the leading streaming platform of eSports related content. Created in November 2018, it still hasn’t 12 months of existence but a lot of learnings in its rear mirror and a lot of future to come in its high beam. Especially regarding Data Management.
To apply to speak or sponsor our 2020 events goto www.digitalenterprisefest.com
Take the Big Data Challenge - Take Advantage of ALL of Your Data 16 Sept 2014pietvz
A customer service call can transform internal processes. Information in Tweets and reviews can lead to better products. Structured and unstructured data brought together can reveal patterns and relationships that unlock powerful business opportunities. We will discuss real-world use cases and best practices for building the infrastructure you need to power Big Data analytics solutions. From the latest in Hadoop innovation, cognitive computing, and cloud-based analytical web services, you will learn how organizations large and small can harness the power of unstructured human information to create, deploy, and deliver the next generation of analytics applications.
Powered by HP IDOL, HP Autonomy delivers intelligent applications that allow your organization to understand the concepts and context of all information in real time, mitigating risk and identifying opportunity. Join us at this session to learn how HP Autonomy can unlock the value of your company’s structured and unstructured data for better insight and greater competitive advantage. HP IDOL, the OS for human information, enables you to index, manage, and process all your data, both structured and unstructured. Learn how HP IDOL delivers unprecedented insights into optimized architecture, scalability, performance, mapped security, and connectivity. Find out more about IDOLOnDemand.com and how you can leverage this revolutionary technology in your own organization.
Bio IT World 2019 - AI For Healthcare - Simon Taylor, LucidworksLucidworks
Presentation from Bio IT World, Boston | April 16-18, 2019
Track: AI for Healthcare: Practical Application of AI in Clinical Healthcare
Session Title: To AI or Not to AI, That Is the Question
Speaker: Simon Taylor, Lucidworks
Designing a Successful Governed Citizen Data Science StrategyDATAVERSITY
To compete in today’s digital economy, enterprises require new ways to expand AI across their entire organization. Nearly all firms want to do more with data science, but they don't know where to begin or how to properly empower citizen data scientists to avoid common AI gone wrong accidents.
In this session, we will discuss how to approach your journey into citizen data science with existing analytics talent. Proven best practices and lessons learned from successful early adopters of augmented data science will be shared. We will walk through example initiative roadmaps, recommended staffing, upskilling, mentoring and ongoing governance.
Microlent System is one of the leading software development company in India.
Microlent System has 9+ Years experience in IT Industry.
we have Industry experts for your business consult IT services solutions.
Services :
Mobile App Development
Web development
IT Services
ERP Development
QA and testing
Software Development Service
Read More :
https://microlent.com/productportfolio.pdf
With the Industry 4.0 technologies, enterprises build digital models of the ongoing processes in the companies, reveal the causes of the current situation, predict future scenarios, and plan changes to be adopted.
91% of industrial enterprises have been already investing in digital factories.
Industrial Internet, Should I be Interested?ionSign Oy
The Industrial Internet of Things - or the Smart Connected Products - are emerging and transforming both competition and companies. What's it all about and why should I care?
SAP Hybris solutions are all about providing a connected front office. But the customer experience can easily get damaged if the data from your business partners or end customers is not secure. With the new EU General Data Protection Regulation (GDPR) coming into effect in May 2018, the need to protect your customers’ data is essential for your business. Learn how to reduce cost by integrating security into your implementation process to be ahead of the curve for future cyberattacks.
Salma Karina Hayat is Conscious Digital Transformation Leader at Kudos | Empowering SMEs via CRM & Digital Automation | Award-Winning Entrepreneur & Philanthropist | Education & Homelessness Advocate
When listening about building new Ventures, Marketplaces ideas are something very frequent. On this session we will discuss reasons why you should stay away from it :P , by sharing real stories and misconceptions around them. If you still insist to go for it however, you will at least get an idea of the important and critical strategies to optimize for success like Product, Business Development & Marketing, Operations :)
Reflect Festival Limassol May 2024.
Michael Economou is an Entrepreneur, with Business & Technology foundations and a passion for Innovation. He is working with his team to launch a new venture – Exyde, an AI powered booking platform for Activities & Experiences, aspiring to revolutionize the way we travel and experience the world. Michael has extensive entrepreneurial experience as the co-founder of Ideas2life, AtYourService as well as Foody, an online delivery platform and one of the most prominent ventures in Cyprus’ digital landscape, acquired by Delivery Hero group in 2019. This journey & experience marks a vast expertise in building and scaling marketplaces, enhancing everyday life through technology and making meaningful impact on local communities, which is what Michael and his team are pursuing doing once more with Exyde www.goExyde.com
Best Crypto Marketing Ideas to Lead Your Project to SuccessIntelisync
In this comprehensive slideshow presentation, we delve into the intricacies of crypto marketing, offering invaluable insights and strategies to propel your project to success in the dynamic cryptocurrency landscape. From understanding market trends to building a robust brand identity, engaging with influencers, and analyzing performance metrics, we cover all aspects essential for effective marketing in the crypto space.
Also Intelisync, our cutting-edge service designed to streamline and optimize your marketing efforts, leveraging data-driven insights and innovative strategies to drive growth and visibility for your project.
With a data-driven approach, transparent communication, and a commitment to excellence, InteliSync is your trusted partner for driving meaningful impact in the fast-paced world of Web3. Contact us today to learn more and embark on a journey to crypto marketing mastery!
Ready to elevate your Web3 project to new heights? Contact InteliSync now and unleash the full potential of your crypto venture!
What You're Going to Learn
- How These 4 Leaks Force You To Work Longer And Harder in order to grow your income… improve just one of these and the impact could be life changing.
- How to SHUT DOWN the revolving door of Income Stagnation… you know, where new sales come into your magazine while at the same time existing sponsors exit.
- How to transform your magazine business by fixing the 4 “DON’Ts”...
#1 LEADS Don’t Book
#2 PROSPECTS Don’t Show
#3 PROSPECTS Don’t Buy
#4 CLIENTS Don’t Stay
- How to identify which leak to fix first so you get the biggest bang for your income.
- Get actionable strategies you can use right away to improve your bookings, sales and retention.
Textile Chemical Brochure - Tradeasia (1).pdfjeffmilton96
Explore Tradeasia’s brochure for eco-friendly textile chemicals. Enhance your textile production with high-quality, sustainable solutions for superior fabric quality.
Create a spend money transaction during bank reconciliation.pdf
Applied AI for Startups
1. APPLIED AI
FOR STARTUPS
Workshop from Adrien HERNANDEZ & Nathalie NERIEC – October 2022
COOPERATHON – The largest Open Innovation challenge in Canada
2. Adrien Hernandez
Data Scientist
Adrien is a data scientist specialized in designing, developing and
deploying scalable real-world machine learning products. He is working
in the largest credit union in North America, where he’s been involved in
the creation and development of innovative AI apps that have enabled
multiple business teams to better meet the needs of more than 7.5
million Canadians.
Having 4+ years of experience in the field of data science across
Canada, the US and France, he started his career in the “French Tech”,
working for a startup that created a mobile application that uses AI
(OCR) to help more than 1 million people in 30+ countries choose better
and safer cosmetic products.
Linkedin: adrienhernandez
Website: adrienhernandez.com
3. Nathalie Neriec
Lead of AI
After her PhD and her postdoctoral position in bioinformatics, Nathalie
has been developing partnerships between industry and academia, with
a focus on advanced analytics. As a business development director at
Mitacs and a lead of AI at Desjardins, she has handled hundreds of
collaborative R&D projects for over 120 companies. She has developed a
strong expertise in the alignment of industrial strategic needs and
academic partners’ priorities, building mutually beneficial and long-
lasting collaborations.
Linkedin: nathalieneriec
4. What we are used to hearing about AI …
(And that doesn’t help)
BLACK
BOX
INPUT OUTPUT
Computer
science
Artificial
Intelligence
Machine
Learning
5. Predictive model
Talent
Management
Re training process
Best practices
Evaluation
Data validation
Cybersecurity
Labelling
Feature engineering Data Protection Business process
integration
Governance
Planning
Data storage
Monitoring
Regulatory
compliance Testing
Reporting & BI
Surveillance
Partnerships
Software engineering
MLOps
Success
measurement
Change
management
Scaling
Maintenance
Documentation
Deployment
Technological
infrastructure
Knowledge
management
What is really AI at a company?
6. Let’s use an example: Meet our startup “IsitSafe.ai”
IsItSafe.ai app
7. What are the business needs of IsItSafe.ai?
IsItSafe.ai app
Users will get a picture of the product from
their phone
We extract information from the picture
We inquire databases to validate regarding:
- HPFB, FDA or EMA approved
- Allergens presence
We return the appropriate information on
the client’s phone in under 1 second
8. Could we use something easier than machine learning?
IsItSafe.ai app
BARCODE INGREDIENTS
9. What are the business needs of IsItSafe.ai?
IsItSafe.ai app
RECEIVES BARCODE
INFORMATION
We return the appropriate
information on the client’s
phone in under 1 second
Product Name Ingredients list
1500004775 ProductName WHEY PROTEIN CONCENTRATE (FROM
MILK, ENZYMATICALLY HYDROLYZED,
REDUCED IN MINERALS), …
Ingredients USA CANADA FRANCE …
WHEY PROTEIN
CONCENTRATE
APPROVED APPROVED APPROVED …
… … … … …
MATCHING PRODUCT INGREDIENTS LIST WITH DATABASE
OF CONTROVERSED INGREDIENTS
10. RECEIVES BARCODE
INFORMATION
What are the business needs of IsItSafe.ai?
IsItSafe.ai app
No need for machine learning!
Issues with barcode approach:
- A barcode database is hard to maintain up to date
- Hard to scale internationally
- Cannot scale up (baby food, etc.)
Product Name Ingredients list
1500004775 ProductName WHEY PROTEIN CONCENTRATE (FROM
MILK, ENZYMATICALLY HYDROLYZED,
REDUCED IN MINERALS), …
11. Could we use something easier than machine learning?
IsItSafe.ai app
BARCODE INGREDIENTS
12. Do I need ML to use photos of “ingredients”?
IsItSafe.ai app
RECEIVES IMAGE
We return the appropriate information on
the client’s phone in under 1 second
Ingredients USA CANADA FRANCE …
WHEY PROTEIN
CONCENTRATE
APPROVED APPROVED APPROVED …
… … … … …
MATCHING PRODUCT INGREDIENTS LIST WITH DATABASE
OF CONTROVERSED INGREDIENTS
OCR (Optical Character Recognition)
Image Text
INGREDIENTS: WHEY PROTEIN CONCENTRATE (FROM
MILK, ENZYMATICALLY HYDROLYZED, REDUCED IN
MINERALS), VEGETABLE OILS (PALM OLEIN, SOY,
COCONUT, HIGH OLEIC SAFFLOWER OR HIGH OLEIC
SUNFLOWER), LACTOSE, CORN MALTODEXTRIN, AND
LESS THAN 2% OF: POTASSIUM HYDROXIDE, CALCIUM
CHLORIDE, POTASSIUM PHOSPHATE, SODIUM
ASCORBATE, SODIUM CITRATE, CHOLINE BITARTRATE, 2’-
0-FUCOSYLLACTOSE*, M.ALPINA OIL**, C.COHNII OIL***, …
AI SOLUTION
13. Do I need ML to use photos of “ingredients”?
IsItSafe.ai app
RECEIVES IMAGE
We return the appropriate information on
the client’s phone in under 1 second
Ingredients USA CANADA FRANCE …
WHEY PROTEIN
CONCENTRATE
APPROVED APPROVED APPROVED …
… … … … …
MATCHING PRODUCT INGREDIENTS LIST WITH DATABASE
OF CONTROVERSED INGREDIENTS
AI SOLUTION
USING AN API IN-HOUSE
CUSTOM TRAINED
MODEL’S WEIGHTS
14. IN-HOUSE
USING AN API
AI SOLUTION
CUSTOM TRAINED
MODEL’S WEIGHTS
“Buying it” “Doing It Yourself”
IsItSafe.ai app
INGREDIENTS: WHEY PROTEIN CONCENTRATE (FROM
MILK, ENZYMATICALLY HYDROLYZED, REDUCED IN
MINERALS), VEGETABLE OILS (PALM OLEIN, SOY,
COCONUT, HIGH OLEIC SAFFLOWER OR HIGH OLEIC
SUNFLOWER), LACTOSE, CORN MALTODEXTRIN, AND
LESS THAN 2% OF: POTASSIUM HYDROXIDE, CALCIUM
CHLORIDE, POTASSIUM PHOSPHATE, SODIUM
ASCORBATE, SODIUM CITRATE, CHOLINE BITARTRATE, 2’-
0-FUCOSYLLACTOSE*, M.ALPINA OIL**, C.COHNII OIL***, …
IsItSafe.ai app
INGREDIENTS: WHEY PROTEIN CONCENTRATE (FROM
MILK, ENZYMATICALLY HYDROLYZED, REDUCED IN
MINERALS), VEGETABLE OILS (PALM OLEIN, SOY,
COCONUT, HIGH OLEIC SAFFLOWER OR HIGH OLEIC
SUNFLOWER), LACTOSE, CORN MALTODEXTRIN, AND
LESS THAN 2% OF: POTASSIUM HYDROXIDE, CALCIUM
CHLORIDE, POTASSIUM PHOSPHATE, SODIUM
ASCORBATE, SODIUM CITRATE, CHOLINE BITARTRATE, 2’-
0-FUCOSYLLACTOSE*, M.ALPINA OIL**, C.COHNII OIL***, …
AI SOLUTION
15. Live Demo: Trying Google’s* OCR API
USING AN API
*Could also work very well on MS Azure, AWS and other platforms.
16. What are the most famous platforms* providing APIs for AI?
USING AN API
*This list is non exhaustive.
17. What are the most common usage of AI APIs?
USING AN API
COMPUTER VISION NATURAL LANGUAGE/SPEECH DECISION
OCR, image recognition, face
detection, object detection, …
Translation, speech to text, text to
speech, speech translation,
speaker recognition, sentiment
analysis, chatbot, …
Fraud detection, forecasting,
anomaly detection, recsys, …
🖥️ 👁️ 🖥️ 🖥️ 📈
‘s GPT-3 model available through an API
18. What are the pros and cons of using an API?
USING AN API
Pros:
• When you start and want to get an MVP out as soon as
possible
• When you’re working on a POC
• Don’t necessarily have in-house expertise in machine
learning – Be careful though
• To accelerate a startup’s path to product-market fit
• To offload the compute and infrastructure challenges of
the AI solution to a larger company
Cons:
• Relying on centralized entities for both training and
inference
• These entities “control” your product destiny
• IP leakage/data leakage
• Cost of goods sold impacted from calling these APIs
• Can become very expensive when you scale
• Sometimes cannot be fine-tuned with your own data
• Model performance can be unclear when used in the
real world
Inspired from How to use massive AI models (like GPT-3) in your startup
19. Shall we use an API for IsItSafe.ai?
IsItSafe.ai app
RECEIVES IMAGE
We return the appropriate information on
the client’s phone in under 1 second
Ingredients USA CANADA FRANCE …
WHEY PROTEIN
CONCENTRATE
APPROVED APPROVED APPROVED …
… … … … …
MATCHING PRODUCT INGREDIENTS LIST WITH DATABASE
OF CONTROVERSED INGREDIENTS
AI SOLUTION
USING AN API IN-HOUSE
CUSTOM TRAINED
MODEL’S WEIGHTS
20. Can we make IsItSafe.ai better with an in-house AI?
After having deployed our MVP using the OCR API from one of the major AI APIs platforms, we could realize that the
model performs poorly for cases where the ingredients list is written on curved products.*
The data scientist could try applying techniques and algorithms to correct the curved effect of pictures before submitting
to the OCR API. However, we would like to take a more robust approach since the business wants to move to other type
of packaging that could be even more curved. (Baby oils, lotions and creams)
*It was the case in 2017, but not anymore. OCR models performance increased tremendously! (thanks Transformers!)
21. IN-HOUSE What is the general outline of a “in-house” AI project?
Source: Full stack deep learning. Lecture 1: Course Vision and When to Use ML
Planning &
Project setup
Data collection &
labeling
Training &
debugging
Deploying &
testing
22. IN-HOUSE What is the general outline of a “in-house” AI project?
Planning & project setup
Source: Full stack deep learning. Lecture 1: Course Vision and When to Use ML
Planning &
Project setup
Data collection &
labeling
Training &
debugging
Deploying &
testing
23. IN-HOUSE What is the general outline of a “in-house” AI project?
Planning &
Project setup
• What are our goals? What problem do we want to solve? Where and how the model is going to be used?
• What about our data? How hard is it to acquire our data? Do we have to label them and how? Is it expensive? How much
data will be needed? What is our data quality? Are there any data security requirements? …
• What about the problem difficulty? Is it feasible? Is it realistic? Is it will defined? Is there any publications on similar works?
What are the computation and technical requirements?
• Accuracy requirement: How costly are wrong predictions? How frequently does the system need to be right tot be useful?
• What are the ethical implications?
• Do we have a team? Do we have the skills?
A data scientist cannot evaluate a specific task without having seen the data
24. IN-HOUSE Important consideration: the lifecycle of the in-house
ML project
MLOps level 0: Manual process
Source: MLOps: Continuous delivery and automation pipelines in machine learning
Planning &
Project setup
25. IN-HOUSE Important consideration: the lifecycle of the in-house
ML project
MLOps level 1: ML Pipeline automation
Source: MLOps: Continuous delivery and automation pipelines in machine learning
Planning &
Project setup
26. IN-HOUSE What is the general outline of a “in-house” AI project?
Data collection & labeling
Source: Full stack deep learning. Lecture 1: Course Vision and When to Use ML
Planning &
Project setup
Data collection &
labeling
Training &
debugging
Deploying &
testing
27. Data collection
& labeling
IN-HOUSE First thing first: Do you have the data?
• What kind of data do we have or want to use? Can we collect is somewhere? Do we want to use public data, or synthetic
data, …? If it is extremely hard to get the data, what do we do?
• In the case where we need to label our data, is it easy to do in our context? If not do we want to spend months on it? If not,
could we reformulate the problem or the business needs in order to make the labeling part feasible? If not, should we
outsource the data labeling part? …
• And by the way, where do we store our data according to our needs? In the cloud? …
Data collection, cleaning, processing and augmentation is one of the most important part of the project
28. Data collection
& labeling
IN-HOUSE Why are good labels so important?
Labeled data is data that comes with a tag, like a name, a type or a number.
Unlabeled data is data that comes with no tag.
It is better to have labeled data. You can do much more with it.
Source: Grokking Machine Learning: what is the difference between labeled and unlabeled?
29. Data collection
& labeling
IN-HOUSE What are the different wats of dealing with data labeling?
Semi-supervised learning
In-house data labeling
Using labeling software
Crowdsourcing
(E.g. Amazon Mechanical Turk)
Data Augmentation
Using synthetic data
Ask your users to label the
data for you
Hiring annotators
Weak supervision
Outsourcing to specialized
companies
Use more advanced
techniques
(self-supervised learning)
30. Data collection
& labeling
IN-HOUSE Some tips about data labeling from FSDL*
Source: Full stack deep learning. Lecture 1: Course Vision and When to Use ML
• You can evaluate your current ml model and see on which cases it is performing poorly to improve your labels on those
specific cases.
• Use labeling software and get to know your data by labeling it yourself for a while
• Write out detailed rules and outsource to full-service company if you can afford it
• Hiring part-time makes more sense that trying to make crowdsourcing work
Don’t forget that garbage in means garbage out
31. Data collection
& labeling
IN-HOUSE How about IsItSafe.ai?
In our context there are several possibilities. We could use a pre-trained model, that has already been trained by somebody else
and then not have to deal with training our own model on our own data.
But its performance could be bad, and thus we would need to fine-tune it with our own data or, even, train our own model from
scratch … So we need data!
We can search for public data online (E.g. EMNIST) or we can create our own synthetic data.
Labels Letters
W
H
E
Extracted list of real words
from FDA website
Synthetized data
creation
Labels Words
ACETAL
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
Data augmentation to face the
problem of curved pictures
Labels
ACETAL
ACETAL
.
.
.
.
.
.
Words
32. Data collection
& labeling
IN-HOUSE Where to find data?
Open source datasets:
Webscrapping data (make sure to check the regulations first):
The internet / social networks
Collecting or creating your own data (synthetized data):
E.g. Tesla collect tremendous amount of data from tesla cars.
Buying data…
33. IN-HOUSE What is the general outline of a “in-house” AI project?
Training & debugging
Source: Full stack deep learning. Lecture 1: Course Vision and When to Use ML
Planning &
Project setup
Data collection &
labeling
Training &
debugging
Deploying &
testing
34. Training &
debugging
IN-HOUSE What about the training phase?
• The training is the phase where we will train our model on our data to perform a specific task.
• A good practice is to create a first simple model called baseline that you will try to improve later with a more complex and
powerful model.
• Don’t hesitate to find an already existing state of the art open-source model and make it work with your data. Don’t reinvent
the wheel!
• Have a dashboard to monitor your model training and serving once used in the real world.
• If the model is performing poorly, it means that the quality of the data could be poor as well. We could need more data or
take a more robust labeling approach.
Honestly, being able to evaluate your model results is even more important than training.
Source: Full stack deep learning. Lecture 1: Course Vision and When to Use ML
35. Training &
debugging
IN-HOUSE What does fine-tuning mean?
Which of the 2 candidates is more likely to perform well when trained to skateboard?
Candidate 1
Candidate 2
It is probably easier to teach a Shiba
to skateboard if he has already been
trained to ride a scooter.
36. Training &
debugging
IN-HOUSE Using foundation and pre-trained models
They allow startups, researchers and others to quickly get up to speed on the latest machine learning approaches without
having to spend the time and resources needed to train these models from scratch. E.g. GPT-3, BERT, DALL-E, …
Be careful of dataset alignment. A pre-trained model trained on the internet data before 2019 won’t know that covid-19 exists!
Source How to use massive AI models (like GPT-3) in your startup
Image source
Fine tuning
37. Training &
debugging
IN-HOUSE Hugging face: Welcome to the world of open source
• 10000+ datasets available
• 75000+ models and pre-trained models available
• Spaces: Share and discover ML apps made by the
community
• A lot of awesome documentation to learn how to
use and deploy state of the art models
• Get helped and use inference APIs
38. IN-HOUSE What is the general outline of a “in-house” AI project?
Deploying & testing
Source: Full stack deep learning. Lecture 1: Course Vision and When to Use ML
Planning &
Project setup
Data collection &
labeling
Training &
debugging
Deploying &
testing
39. Deploying &
testing
IN-HOUSE Putting model into production to see if it really works
• It is very important to get a MVP and building a prototype as early as possible! You will realize that in the real world it may
not work as well as in your development environment. Is your model quick enough in terms of predictions speed? Will your
ml product scale well when the number of users will increase? Will your ml product be easily maintainable?
• Have a simple interface where users/beta testers can try your ml product and provide you with feedback.
• It is also recommended to use your model as a service. You don’t want to have everything in the same script/environment
(UI, model, data, …). Learn about the concepts of REST APIs, micro-services, …
• Do you have good success criteria, and are they met when the model is deployed for real?
• You could realize that the model’s performance isn’t great and that the model is slow to produce results (inference time)
when used in the real world! You may not need business requirements! What should you change, revisit?
What will make your product successful is not spending months doing research to apply latest SOTA models to get a 3% gain in
your prediction quality. It is to have a ml product that is scalable and maintainable.
Source: Full stack deep learning. Lecture 1: Course Vision and When to Use ML
40. IN-HOUSE Extra consideration for all AI projects
Source: Full stack deep learning. Lecture 1: Course Vision and When to Use ML
Planning &
Project setup
Data collection &
labeling
Training &
debugging
Deploying &
testing
Infrastructure and tools
Having the right team
Ethics
41. IN-HOUSE
Source: Full stack deep learning. Lecture 2: Development Infrastructure & Tooling
Infrastructure and tools
42. Having the right team
IN-HOUSE
Source: Full stack deep learning. Lecture 8: ML Teams and Project Management.
Role Job Function Work Product
ML product manager Work with ML team, business,
users, data owners to prioritize
& execute projects
Design docs, wireframes, work
plans
MLOPs / ML Platform Build the infrastructure to make
models easier to deploy, more
scalable, etc
ML infrastructure
ML Engineer Train, deploy & maintain
prediction models
Prediction systems running on
real data in production
ML Researcher Train prediction models (often
forward looking or not
production-critical)
Prediction model & report
describing it
Data Scientist Blanket term used to describe
all of the above. In some orgs,
means answering business
questions using analytics
Prediction model or report
ML talent is expensive and scarce; ML projects have unclear timelines and uncertainty; ML can lead to technical debt
43. Having the right team
IN-HOUSE
MACHINE LEARNING ENGINEER
THE NEW UNICORN?
45. In conclusion
IsItSafe.ai app
AI SOLUTION
USING AN API IN-HOUSE
CUSTOM TRAINED
MODEL’S WEIGHTS
Planning &
Project setup
Data collection
& labeling
Training &
debugging
Deploying &
testing
Infrastructure and tools
Having the right team
Ethics
46. THANK
YOU!
Workshop from Adrien HERNANDEZ & Nathalie NERIEC – October 2022
COOPERATHON – The largest Open Innovation challenge in Canada