Tuesday 29 June 2021

Data Science - The beating heart of AI

Another conference produced at the Adastral Park Hybrid studio and this time we were able to invite young researchers on site to watch the production of the conference live "Top of the Pops - style"!



TFNetwork teamed up with iSee 
VIRTUAL
- Conference & Workshop -

Workshop: iSee - What do Users Want when Explaining an AI System?
An Introduction to the iSee Project and Co-Creation Activities.
Find out more about iSee on our iSee workshop page.


Conference registration is closed now
Please find recordings below with the respective speaker profile
or

Conference: Data Science - The beating heart of AI
All of the recent successes in Artificial Intelligence (AI) have been made possible by analysing data – lots of it.

Data is the raw material for the type of machine learning that has been so successful in creating ever more AI applications. Data Science is at the heart of it. It sits at the cross-roads of statistics and machine learning. Data Science turns data into insights which can be cast into models through machine learning. Over three days we will explore the state of the art in Data Science and how it leads to applications in industry.

🚩Tuesday 12th October 2021, 10:00 - 12:30
All models are wrong! – The foundations of modern Data Science

10:00 - 10:15    Prof Tim Whitley, MD Applied Research for BT and MD of BT’s Technology Campus ‘Adastral Park’
Welcome to 'Data Science - The beating heart of AI'
Recording available, please click link above

10:15 - 10:30    Dr Detlef Nauck, Head of AI & Data Science Research at BT, Visiting Professor at Bournemouth University
Programming with Data
Experienced Data Scientists know that the biggest challenge to using insights from data in an operational setting is to get hold of good quality data. Pretty much every talk or blog mentions that 80% of the effort in any data science project are spent on data access and data wrangling. My take is that this effort will approach 99% soon simply because the analytics and machine learning parts are becoming largely automated. It is time that we shift focus from techniques and software to data and make it clear to ourselves what we are doing in Data Science and in particular in Machine Learning – we are programming with data.
Recording available, please click link above

10:30 - 10:45    Michael Free, BT Michael Free - AI Research Manager
Models that cheat - Making sure it really works
Machine Learning has made great strides in hard, unstructured problems with the advent of deep learning. However, such progress does not come free of issues. Often treated as black box solutions, interpretability and explainability are complex issues when building deep learning models, and poorly framed experiments and ‘dodgy data’ have led to a litany of models that don’t really work in practise – they merely ‘cheat’ the limited test you’ve given them.
Recording available, please click link above

10:45 - 11:00 Blaise F Egan, BT Data Science and Statistics Specialist
Statistics: the original data science
Statistical science is the oldest of the components of what is now called Data Science. I will give a quick run-through of some of the bigger landmarks on the road to where we are now and how we got here.
Recording available, please click link above

11:00 - 11:15 Rob Claxton, BT Senior Manager – Big Data, Insight & Analytics
Are we nearly there yet? Model baselines and performance
The process of building models has a strong emphasis on ‘digital performance testing’. In other words, how accurate is my model? Whilst this makes sense when building prototypes or testing the state of the art, it can lead to problems when developing models for production. In this talk we turn the tables and ask the question, how good does my model need to be?
Recording available, please click link above

11:15 - 11:30    Prof Alessio Lomuscio, Department of Computing, Imperial College London
Towards verifying neural systems
A key difficulty in the deployment of AI solutions, including machine learning, remains their inherent fragility and difficulty of certification and explainability. Formal verification has long been employed in the analysis and debugging of traditional computer systems, including hardware and networks, but its deployment in the context of AI-systems remains largely unexplored.
Recording available, please click link above

11:30 - 11:45    Dr Raoul-Gabriel Urma, CEO Cambridge Spark
Skills for a Data World
Data is all around us and transforming how organisations are having to operate. As a result, data is significantly impacting the future of work. In this talk, we will cover what are the key skills for individuals and organisation to invest in across all roles for a successful future.
Recording available, please click link above

11:45 - 12:30
Panel on 'Foundations of Data Science' and Q&A
When we analyse data we often search for a model to describe what we find, to make decisions, or to make predictions. A model is always a simplification of what we find in the data and so “all models are wrong but some are useful” (George Box, 1978). What do we need to do to make sure our models are as good and useful as they can be?



🚩Wednesday 13th October 2021, 14:00 - 16:30
That doesn't look right! – How to find the glitches in your data and models (anomaly detection, bias/fairness)

14:00 - 14:10    Julien Gruhier, Senior Manager Data Scientist at BT
Challenges of analysing mobile network data
The Geo-spatial analytics market is large and expanding every year. As a Telecommunication operator we can access part of this market relying primarily on the meta data of our mobile network. Despite a limited location accuracy compared with GPS based providers, we are nevertheless a strong contender. However, there are numerous data science challenges that need to be tackled to extract relevant insights from this wealth of information. I will touch on some of our core challenges to filter out noise from the signal to enable the extraction of data insights and help our customers to make data driven decisions.
No recording available

14:10 - 14:15    Oliver Waring, Senior Data Science Specialist at BT
Q&A for "Challenges of analysing mobile network data"
No recording available

14:15 - 14:30    Trevor Burbridge, BT
Anomaly Detection
A look at anomaly detection, focussing on streamed data and an examination of the underlying models and potential pitfalls
No recording available

14:30 - 14:45    Kes Ward, Mathematics PhD student at Lancaster University
Anomaly Detection at the Edge
In an Internet of Things where everything is collecting and analysing its own data, we need edge analytics to help us sort the meaningful from the muck without breaking the computational bank. In this talk I will present a new statistical method for finding anomalies of different shapes and sizes in a real-time data signal, while working under extremely tight computational constraints.
Recording available, please click link above

14:45 - 15:00    Faisal Nazir and Subhash Talluri,  AWS at Amazon Web Services (AWS)
Faisal Nazir on "The importance of explainability of AI"
Data scientist could potentially wield great power over the lives of everyday people. This power comes from how they develop ML models that can be used to make life-changing decisions. Explainability - having knowledge of why an model makes an inference - is the field that tries make sense of a models decision. We will discuss what tooling is available to Data Scientists to help them find out what is going on with the models they train.
Subhash Talluri on "Data Science on AWS"
AWS has been continually expanding its service portfolio to support virtually any cloud workload, including many services and features in the area of artificial intelligence. In the context of data science projects on AWS, the benefits of cloud computing include agility, cost savings, elasticity, faster innovation and smooth transition from prototype to production. Amazon SageMaker is a fully managed offering that addresses every aspect of machine learning by its modular design. All machine intelligence is powered by data. However, not all data are created equal. We need to critically evaluate machine-learning products from a standpoint that prioritizes the quality of the data streaming into them. This necessitates the need for a data lake or a data platform with considerations
Recording available, please click either of the links above

15:00 - 15:15    Ben Taylor, CTO Rainbird
Bias, transparency and governance in automated decision making
As they look at the great landscape of AI, organisations are getting to picture machine learning in finer detail. But the closer they get to the detail, the more they notice a chasm emerging between prediction and automated decision. And for no organisation does that chasm pose greater danger than those who operate in regulated industries.
Recording available, please click link above

15:15 - 15:30  Alex Healing, Future Cyber Defence Research at BT
Human-Machine Collaborative Analytics
The use of AI is clearly a critical part of analysing data at the scale that today’s IT systems allow, but sadly the human user in the system is often an afterthought. This talk will touch on the challenge, opportunity and progress made so far to create more human-centric AI systems for data analysis, involving visual interfaces for presenting machine learning results, and to help analysts better explore and generate insight from data. 
No recording available

15:30 - 15:45 Prof Sam Madden, MIT CSAIL
Outlier Detection
Rapidly developing areas of information technology are generating massive amounts of data. Human errors, sensor failures, and other unforeseen circumstances unfortunately tend to undermine the quality and consistency of these datasets by introducing outliers -- data points that exhibit surprising behaviour when compared to the rest of the data.
Recording available, please click link above

15:45 - 16:30
Panel on 'Data quality and data anomalies' and Q&A
Sometimes the data we work with is not right. Some parts may be missing or some may be plain wrong. Data can misrepresent the world which leads to bias in models. How do we find what’s wrong in data and models – or what looks wrong but tells us something interesting?



🚩Thursday 14th October 2021, 10:00 - 12:30
Where's my model? – Putting it all together and running the AI factory

10:00 - 10:15    Zoë Webster, Artificial Intelligence Director - Data Solutions, BT
Establishing an AI Centre of Enablement
There are a number of questions to consider when setting up a team to accelerate the development and adoption of AI at scale. Not all of these centre on the technology. This talk will discuss some of the questions that have focused my attention here at BT and the perspectives that are important in addressing them.
Recording available, please click link above

10:15 - 10:30    James Hamilton, TUI Head of Commercial Analytics
It’s not all about the model – challenges applying artificial intelligence
In TUI we have a very successful automated pricing system that has delivered significant benefits over a number of years. We continue to develop it incrementally and our current focus is on better use of our data and applying artificial intelligence. We have experienced a number of challenges in applying these approaches and I will talk about our experiences, how we are overcoming them and why if we get to training a new model we have already done the hard part.
No recording available

10:30 - 10:45    Tuba Islam, Google Cloud
ML Journey into production - What are the challenges and how to tackle them?
In this talk, we will share the key challenges that the organizations are likely to face when they move their machine learning models from experimentation stage into production and provide recommendations on how to handle these challenges with the right process and technology in place.
No recording available

10:45 - 11:00  Prof Maria Fasli, Executive Dean, Faculty of Science and Health, University of Essex
Collaboration in Data Science Projects
No recording available

11:00 - 11:15    Libby Kinsey, Head of Data Science - Strategy and Operations at Ocado Technologies
Scaling Data Science
Data science already powers insights, products and capabilities in every part of Ocado Technology’s solutions, i.e. we know how to do data science at scale. The next big challenge is about reducing effort to value. Scaling data science is hard.
Recording available, please click link above

11:15 - 11:30    Dr James Grant, Lecturer in Statistics, Lancaster University
Multi-armed Bandits
In modern data science applications, there is often the opportunity to observe the effects of a decision and revise it, and to iterate this process repeatedly, experimenting in order to learn an optimal action. Multi-armed bandits provide mathematical models of such settings, where designing an optimal sequence of decisions can be highly challenging. This talk will give an introduction to multi-armed bandit models, and explore the best techniques used to tackle the problems.
Recording available, please click link above

11:30 - 11:45    Dr Matloob Khushi, Senior Lecturer in AI at the University of Suffolk - UoS
AI: The Fourth Edge
In this talk Matloob will talk about how AI has transformed our society and has defined new norms of livings. He will also share some of the outcomes of his AI research.
Recording available, please click link above

11:45 - 12:30
Panel on 'Running Data Science for real' 
When we have analysed all the data, when we have built all the models, we need to bring everything back from the lab to the real world. How does Data Science ultimately lead to models in operation and how do we keep track of it all?




Watch the highlights of the last conference produced at that Adastral Park Hybrid Studio

Catch up on the presentations and panel discussions on YouTube

Posted by
Labels: ,

No comments:

Post a Comment

Note: only a member of this blog may post a comment.

Subscribe to: Post Comments (Atom)