I had the pleasure of attending the 2nd Applying Artificial Intelligence and Deep Learning for Enterprises Conference by Clariden Global. There were many takeaways from the two-day conference and here's a summary:
On data culture
- There was a lot of emphasis on building the data culture within the organisation.
- People, process, technology (or architecture) are the typical three main pillars organisations need to focus on to create a data-driven enterprise. One end-goal would be to empower self-service with a high level of automation, and this can begin with getting people to identify the business problem(s) and think cross-function/ industry.
- Readiness challenges often include skills gap and the lack of resources to analyze data, failure to identify AI use cases, the inability to define an AI strategy and budget constraints.
- Strong change management process is necessary and it is important to identify a good sponsor as well as active users.
- While analytics can be used for decision support, knowledge discovery and optimization, the more important question is what do we want to prioritize.
- Robots will be helping in supporting elderly as caregivers as well as in the form of a customer service associate.
- A robot can be achieved with the following set-up:
(i) Perception: 3D camera for face recognition, gesture recognition, understanding of social situations; microphone for speech recognition; (ii) Processing/ decision: internal processing for emotion model, memory model, social attention and chatbot; (iii) Action: emotional expression, lip synchronization, online gaze generation, speech synthesis and gestures.
- In the backend, defined Q&A with access to weather channel and other APIs can be loaded and further data about the person/ interaction would subsequently be added to database after each interaction to help enhance the performance of the robot. However, no robots/ machine in the world know what they said/ are saying, as there is no sense of awareness.
On transportation/ logistics management
- Price, time and quality are three aspects that companies in this industry focus on.
- Supply and demand prediction models would need to take into account trends across days of the week, weather, and holidays. In some cases, an ARMA exogenous variable model (ARMAX) could be used to quantify the impact of X on the demand. With the high volume and fast moving nature of the industry, these models need to be able to predict in intervals of 5-60 minutes.
- In order to optimize completion rates/ matching, knowing drivers' preferences improves allocation.
- Route optimization/ traffic prediction is also done and algorithms such as CNN can be used for road networks and building extraction from satellites, while bayesian networks can be used to predict route and LSTM to predict estimated travel time and estimated delivery time.
- Some KPI metrics for the industry include unmet demand and delays.
On computer vision
- Looking for objects is difficult for humans. However the process can be made more efficient and price with AI.
- Automating object detection in real-time drone video stream is useful for monitoring carpark utilization (counting cars)/ crowd (counting people), detecting infrastructure defects (highlighting cracks) and finding survivors (highlighting people), in addition to the most common use case of surveillance.
- In order to perform the deep learning, there are five things we need to have in place/ take note of: (i) neural network architecture, (ii) pre-trained model, (iii) dataset, (iv) GPU computing (training and prediction), (v) model performance i.e. tradeoff of accuracy vs speed. Ultimately, we need to strike a balance between time and false positive/ negatives.
- One challenge in the use of drone in objective detection for navigation is when the model doesn't work from a distance, i.e. the drone is unable to correctly identify the object when it is far away from the object and this requires a lot of re-training.
- For cases where cameras are used for object detection/ centering, we can compare frame vs frame and look at the changes in rgb value; for which, we won't need a neural network.
- Object detection is also especially useful in identifying documents and photo fraud in cases requiring NLP-based document and photo validation. It would be operationally challenging to validate all data and photo manually.
- Chart data, medication, treatment and lab info/ data (readings/ measurements from devices) can be used in prediction models to predict re-admission rate and then compared against healthcare workers' assessment for measuring model performance.
- Bed occupancy rates in hospitals can also be estimated with the use of data.
- With a shift in focus from 'treating disease' to 'regaining wellness', more value-based healthcare models can be developed.
On reducing cost
- Failure demand is costly. An example of failure demand is when customers call in to enquire on status of their claim. This is opposite of value demand where customers call in to perform a transaction. Diagnostic analysis can be done to understand root cause of failure. - With the use of AI, business rules can also be built into production to automatically assess and route to necessary departments to process claims. This also helps in creating consistency in customer experiences.
On predicting user behaviour
- Experiments can be run to better understand user behaviour and preferences. Content/ advertisements are then customized to users more effectively with the right audience targeted/ re-targeted. Additional revenue streams can be identified through testing and learning the market.
- In the context of shopping, likelihood of purchase is predicted based on customers' demographics and transactions (through point-of-sale data, and web/ app history) to develop recommendations on products and stores for each customer.
Here's also appending the agenda for the two-day conference, for reference:
- The AI-First Enterprise: How Automated Machine Learning is used to Transform the Enterprise
by Sebastian Wedeniwski, Chief Technology Strategist, Standard Chartered Bank
- The Intersect of Artificial Intelligence with Robotics: Latest Advances
by Prof Nadia Thalmann, Director, Institute for Media Innovation, Nanyang Technological University
- Advanced AI Experience from Abroad, with any case studies and latest AI innovations
by Lian Jye Su, Principal Analyst, ABI Research
- Operationalising AI
by Vince Kasten, Regional Operations - Robotics and AI, Prudential Corporation Asia
- Grab's Transformational AI Lab in Singapore: How Grab is Using AI to Build its Future Business Platform
by Jagan Varadarajan, Head of Data Science (Machine Learning and Maps), Grab
- Panel Discussion: Getting Ready for AI: How to Integrate AI into Your Existing Infrastructure
Moderator: Jake Saunders, Vice President, Asia-Pacific & Advisory Services, ABI Research
Panelists: Vince Kasten, Regional Operations - Robotics and AI, Prudential Corporation Asia
Doan Do Quy, Chief Digital Officer & Chief Technical Architect, BNP Paribas
Dr Jagan Varadarajan, Head of Data Science (Machine Learning and Maps), Grab
Andrew Koh, Deputy General Manager and Regional Head Risk, Habib Bank Limited
Sebastian Wedeniwski, Chief Technology Strategist, Standard Chartered Bank
- Roundtable Discussion: How Deep Learning and Machine Learning is Transforming Different Vertical Industries Forecasting
Roundtable 1: AI for Power Generation, Transmission, Distribution and Renewables
by Praveen Lala, Regional Director, GE Digital
Roundtable 2: The Use and Implications of AI for Creative Fields Such as Music
by Kat Agres, Research Scientist, Social & Cognitive Computing Department, (IHPC), A*STAR
Roundtable 3: Air Asia Case Study: Achieving Transformative Business Outcomes via AI
by Niyati Goel, Global Rewards CoE Lead, Air Asia
- Transforming Businesses with Artificial Intelligence
by Daniel Sparing, PHD, Machine Learning Specialist, Japan Asia Pacific, Google
- AI in broadcasting: Different Usages in the Industry
by Lisandro Tapia, Technical Design Director, Mediacorp Pte Ltd
- Augmenting Drones with AI
by Jiin Joo Ong, Co-Founder and Chief Technology Officer, Garuda Robotics
- AI for Perception and Tracking, Use-case and Methodology
by Eric Juliani, Head of Deep Learning Team, Easy Mile
- Org 2.0 - Collaborative AI Governance for the Future of Work
by Jason Sosa, CEO & Founder, Blackbox AI
- A Human and Machine Collaborative Future for Research, Innovation and Enterprise (RIE)
by Dr Lua Eng Keong, Director, Intelligent Computing Labs , IMDA; Adjunct Professor, National University of Singapore
- Data-driven Approaches in a Technology Startup
by Michal Szczecinski, Head of Data, GOGOVAN
- Moving AI Off Your Product Roadmap and Into Your Products
by Juliana Chua, Head Digital Transformation, NTUC Income
- Data Analytics and Data Analytics based on AI Models
by Sudhir Panda, Global Data Intelligence Lead, Digital Banking
- AI in Action: Case Study of How AI is Helping Doctors Predict Outcomes of Patients Effectively
by Andy Tan, Principal Medical Informatics Specialist, National University Health System
- AI and Machine Learning - Contributing Across The Spectrum of Healthcare
by Dr Shailendra Bajpai, Head of Disease Management and Stakeholder Engagement, Sanofi
- Utilizing Data in the Newsroom Amidst an Aggressive Digital Transformation at a Century-old Paper
by Romain Rouquier, Asst. Director of Data Analytics, South China Morning Post
- Operationalizing Machine Learning: How to Ensure Value-Driven Deployment
by Praveen Lala, Regional Director, APAC and India, GE Power Digital
- AI for Banking and Financial Services
by Girish Sundaram, Technology Director- Data Science | Analytics | AI, Standard Chartered
- AI for Customer Experience and E-commerce: How ML is Helping NTUC Link Transform a Legacy Loyalty Program into Omni-Channel Personalized Customer Experiences
by Kevin Oh, VP, Head of Customer and Digital Analytics, NTUC Link
- The Artificially Intelligent Fund - What Works
by Victor Lye, Founder & CEO, PIVOT Fintech