Multi-Step Learning and Optimisation

Lead Research Organisation: University of Cambridge
Department Name: Chemistry


Year 1: Generic training activities for all first-year student members of the CDT.

Year 2-4: The quest for effective methods of exploring chemical search spaces has ventured from purely in-silico techniques to cyber-physical ones, guiding either robotic or human experimentalists. This blending of AI with physical systems is becoming a well-established technique, and with the addition of high-throughput chemistry techniques, offers a means to efficiently achieve effective and efficient exploration, potentially exceeding human guided techniques. These methods may also be adapted to become a technique for optimising chemical synthesis problems, targeting certain characteristics and molecular traits. Closed-loop systems have already been developed to design efficient production of target molecules and have found a great deal of success. Some of the difficulties with these techniques come about due to the multiple stages involved in complex chemical synthesis tasks and the plethora of objective functions that must be balanced.

Multi-step modelling and design involves multiple steps forming a chain of models, often appearing in situations where multiple chemical products must be produced, and subsequently reacted, to form a final product(s). Although found extensively in chemical process synthesis and synthetic chemistry, many other similar (and dissimilar) contexts exist. Many of the current methods for multistage modelling rely heavily on expert judgement and guesswork to balance objectives, both inter and intra-stage. This represents an obvious weak link in process synthesis. The removal of this form of guesswork is likely to lead to significant improvement in many measures of process success and a speed-up of development.


10 25 50

Studentship Projects

Project Reference Relationship Related To Start End Student Name
EP/S024220/1 31/05/2019 30/11/2027
2276523 Studentship EP/S024220/1 30/09/2019 29/09/2023 Rory Moncur Geeson
Description Early stages of the award demonstrated the application of semantically enriched knowledge management to combustion data management and curation. This presents a route towards a shift in the workflow of combustion researchers towards more actively curated stores of data and automated screening of experimental results with existing data.
Following on from this strand, this award has identified a more generic formulation for many real-world problems (multi-step problems) and has begun to explore new solution techniques.
Exploitation Route The early stage, knowledge graph based outcomes are already being taken forward within the World Avatar project. Later stage outcomes present many opportunities for exploration of further solution techniques and applications to alternative fields and topics.
Sectors Chemicals,Manufacturing, including Industrial Biotechology,Pharmaceuticals and Medical Biotechnology