KNIME Fall Summit 2020

Speaker: Michael Berthold (KNIME)

KNIME’s unique solution for closing the gap between creating and productionizing data science is Integrated Deployment. It ensures that the entire creation process can be deployed and updated - which is managed by a single platform. This unique functionality eliminates the translation process that emerges in typical set ups. See how this is applied in data science practice.

How do the small banks service their customers nowadays and compete with the megabanks?

• Requirements: A tool that deals better with improving model accuracy and deployment of the model that can scale
• Challenges: Who owns the whole ML pipeline? How can the downstream team embed the selected model into the software? What happens when you need to retrain the model?
• Set-up: KNIME Server and KNIME Integrated Deployment
• Results: Model development and deployment of major machine learning projects - recommender system for NII products, automatic transaction classification, rewards validation at end of each cycle, and successful anomaly detection

Speaker: Zehra Hussein (Kasasa), Moderator: Phil Winters (KNIME)

Speaker: Edgar Osuna (TODO1), Moderator: Jim Falgout (KNIME).

TODO1 built an anti-fraud profiler to deal with transactional fraud mitigation in the digital interactions between financial institutions and their customers.

• Requirements: One platform to accomplish everything; REST API; self-documenting; a good price
• Challenges: Fraud data is hard to get, real-time responses are hard to achieve (<100ms), and downtime was not an option
• Set-up: Running load balancer / 4 x KNIME Server Large / Rabbit ML / 2 x KNIME Executors
• Results: With the system in production in the cloud and fully operational for the past 3 months it has been managing 650+million transactions with no downtime. Overall improvement of +100% fraud detection rates. Round trip median time of approx. 100 ms. That’s impressive!

Speakers: Dean Abbott (SmarterHQ), John Elder (Elder Research), Moderator: Michael Berthold (KNIME)

• Change management, when reality changes and your models no longer work: How do you figure out when a model needs to be changed? Is the problem even a technical problem? Maybe it’s about trust: People fear change. How do we address that?
• New techniques on the horizon: What will come after all the “deeps”? Deep learning is so 1990s! Interpretability, provable accuracy, blackbox vs gray box, and more.
• ML models and bias: Building models that don’t discriminate? But they are supposed to! The analyst’s job is to discover the truth in the data. Deciding what to do with that information is the second step.
• The future of AI: Will it rescue or destroy humanity? Is AI simply good engineering? Will AI exceed human intelligence one day? Or is your fast, loyal, and obedient AI model actually a good German Shepherd dog who will go get that ball?

Speaker: James Grimes (Truata), Moderator: Cynthia Padilla (KNIME)

The goal in this project was to anonymize and automate credit card data and enable financial services companies to perform their own “self-service” analytics while ensuring privacy is maintained.

• Requirements: A tool that is open source, extendible, and that offers Guided Analytics and blueprints as they didn’t want to write their own workflows from scratch. KNIME was a good fit
• Challenges: Authorization in an integrated solutions architecture has serious authentication challenges!
• Set-up: Multiple KNIME Executors and KNIME WebPortal
• Results: Several hundred financial institutions can now access the solution through the WebPortal

Speakers: Danilo Nato (BASF), Roberto Caffaor (Hupdata), Moderator: Cynthia Padilla (KNIME)

How a smaller business (Hupdata) was able to help a large company (BASF Brazil) reach their goals. A new regulatory law in Brazil aims to protect the dangers of exposure to agrochemicals.

• Requirements: 1) Build a RIM (Regulatory Information Management) platform to have all the data in a single location ready to be analyzed. 2) Calculate the risk assessment over 116 BASF products and visualize results interactively to enable risk toxicology experts to see results.
• Challenges: Authorization in an integrated solutions architecture has serious authentication challenges!
• Set-up: KNIME + WebPortal + Guided Analytics + Tableau
• Results: Risk toxicology experts are empowered with technology, saving thousands of hours of work; all numbers are crunched into automated reports. And the solution is scalable. Global roll-out is planned for Europe, Asia and North America.

Speakers: Arturo Boquin (Dinant) & Ignacio Pérez (IQuartil), Moderator: Cynthia Padilla (KNIME)

Hear about Dinant’s new system that produces product quality reports in real time, enabling data driven decisions. Dinant produces snacks, home and personal care products, grease, and oils.

• Requirements: Automate and reduce time taken for what is essentially a simple process of capturing, processing data and deploying insight.
• Challenges: Real-time processing and storage, and change management
• Set-up: KNIME Server - Storage in Datalake - AWS - Deployment to Tableau Server/Dashboard
• Results: Savings in production costs and massive time savings. The standard process took 3 hours and 15 minutes. The new process reduced this time by 89% - down to 21 minutes!

Speaker: Kenneth Longo (Wave Life Sciences), Moderator: Cynthia Padilla (KNIME)

How a drug discovery company that designs short DNA and RNA sequences to treat life threatening disease gets the right information to the right people in a timely manner.

• Challenge #1: Quickly and reliably develop, test, and deploy software that uses modern data science tools and answers key business questions.
• Challenge #2: A platform that provides end to end software development tools for data science backend, data connectivity, UI, UX, and reporting.
• Set-up: KNIME Analytics Platform, KNIME Server, and KNIME WebPortal
• Results: A means to productionizing data science tools for broad use and delivering solutions via the browser.

The journey from using a complicated system to a clean design. Before the change their setup, based around Hadoop, brought with it code promotion issues, which were hard to monitor and required time-based scheduling meaning that deploying a workflow and finding an available slot was tedious work.

• Requirements: New system would be based on Big Query, use a CI/CD process, enable workflow orchestration, and real-time monitoring.
• Set-up: KNIME was put at the centre of the architecture. The clean design of GCP + KNIME + Microservices enables a bridge between the centrally stored data and business applications.
• Results: GDPR is ensured, data quality checks and logging are built into all database services and logging framework - letting users focus on developing new things Notifications received via slack or email.

Speakers: Phil Winters & Paolo Tamagnini (KNIME)

The first half of this presentation discusses the different stakeholders in data science and shows how KNIME Analytics Platform caters to them all by surfacing the capabilities the specific stakeholder requires: team leaders, data engineers/data scientists, business analysts, model/ML operations, IT operations, and compliance. 

• Highlights for Teams: KNIME WebPortal for Interaction, Components, KNIME Hub Public & Private Spaces, Integrated Deployment.
• Highlights for IT & Compliance: Scalability and control, centralized resources / strategies and security plus financial / risk oversight, costs allocation, governance, traceability, GDPR

by Clarkston Consulting

by Day5 Analytics

by IQuartil

by Theobald Software

by Competitive Analytics

by HupData