Hi all,
Tomorrow James will give group meeting. See below for the title and
abstract.
Please note the nonstandard time due to Martin's thesis defense, which will
be in the Division Room afterwards at 4:00 PM (best of luck!!).
Also, Jhonathan, Ben, and I will be hosting the A-G group open house on
October 28th - all are welcome! Each subgroup (quantum, excitonics,
materials), please get together and decide on two posters to submit. More
to follow later.
Best,
Ian
-------------------------------------
Speaker: James (Sung-Jin) Kim
Title: Very high-throughput screening for flow battery molecules
Abstract: A binary kernel machine with an ensemble descriptor (BIKE) is
proposed. BIKE is a prediction method for very high throughput molecular
screening. Using a kernel machine with binary radial basis function, the
severe irregularity between structure and property can be effectively
exploited compared to linear methods. Moreover, an ensemble descriptor
significantly boast the regression performance where different type
fingerprints are simultaneously embedded in the descriptor. Adding a
property specific descriptor, molecular weight, BIKE eventually achieves
R^2 = 0.9 and R^2 = 0.94 for solubility regression and classification,
respectively. For its application of flow battery molecular screening, we
find that correction of the experimental solubility data can improve
estimation accuracy especially for molecules enumerated with strong acid
group such as sulfonic acid.
Show replies by date