the digital transformation of [email protected]...
TRANSCRIPT
The digital transformation of industrial sectorsNikos Kazazakis | Octeract | [email protected]
What is digital transformation?
Physical system Decision
What is digital transformation?
Physical system DecisionDigital
tech
Who takes on the risk of R&D?
Uncertainty space
Startup vs business
Problem Solution
Business
● Fast iterations● Innovation● Efficiency
What is mathematical optimisation?
What is mathematical optimisation?
Physical system DecisionDigital
tech
Modelling interfaces
Data collection
Data analytics OptimisationMachine
learning Simulation
Customer profile
Source: https://discovery.hgdata.com/product/ibm-ilog
Case study: HEN for a large chemical company
● Problem definition: what is the optimal network of heat exchangers to minimise annual operational cost and annualised investment cost?
● Design that cannot be produced by humans (more HEs)● Optimal network improvement: 24%, 47%, ~$5m/year/site
Other applications
● Tailored chemical compounds (17%)● Process control (safety by design)● Operations scheduling (38%)
Improvement by sector
● Any company with more than 50 employees that does not use optimisation is losing money
● Penetration rate <1%● Technological and
knowledge barriers
User workflow
Modelling interface
Do the math Solver Decision
BM ABM
Physical problem
1000s
End goal
DecisionPhysical problem
Modelling interface
Products
Octeract Engine
● Solves any type of optimisation problem● Highest solve rate● Massively parallel● Powered by Reformulator technology
A solution…
the best solution
Products
Octeract Engine
● Solves any type of optimisation problem● Highest solve rate in the world● Massively parallel● Powered by Reformulator technology
Octeract Reformulator
● World’s first optimisation compiler● Produces math automatically● Accelerates workflow by 1000x● Works with any solver● Allows users to save, reuse, and share
mathematical techniques
Reformulation becomes reusable
ORL (10μs)
Manual (5 hours)
Model A Model B
What is “AI”?
Large-scale optimisation to produce a model directly from data● Should be used:
○ Physics uncertain/unknown○ Large volume of data is available○ Transferability is not necessary○ To describe parts of a complex system○ When it’s ok to violate constraints
● Should not be used:○ Physics well understood○ Not enough (relevant) data○ Transferability is necessary○ As a substitute for old-fashioned engineering○ Safety-critical applications
Data challenges
● Hype● Premature collection
○ Noise○ Data cleaning○ Lack of context
● No standardisation● Low transferability● Human vs machine
○ 10x-100x less expensive○ Little to no data collection○ Transferable solutions
Hurdles to digital transformation
● Lead times● Dilution of responsibility● Lack of incentive at lower levels● Initiating communication● Ill-defined innovation budget● Risk aversion● Loss of interest (path of least resistance)● Money is (mostly) irrelevant for tech startups
Solutions
● Lead times● Dilution of responsibility● Lack of incentive at lower levels● Initiating communication● Ill-defined innovation budget● Risk aversion● Loss of interest (path of least resistance)
● Milestone based contracts● Accountability● Rewards/promotion opportunities● Promote outreach● Well-defined innovation budget/rules● Risk sharing
Thank you!