[행사/세미나] [Seminar] May. 12(Fri.) Compressed sensing provides a novel perspective on Drosophila olfactory processing
- 물리학과
- 조회수2493
- 2023-05-10
생물물리계산연구실에서 고등과학원의 최기리 박사님을 초청하여 아래와 같이 세미나를 개최합니다.
학과 구성원 여러분의 많은 참여를 부탁 드립니다.
=============================================================
○ Topic: Compressed sensing provides a novel perspective on Drosophila olfactory processing
○ Speaker: 최 기 리 박사님 (고등과학원)
○ Date/time: 2023. 5. 12. (Fri.) 11:00 AM
○ Place: Natural Science 1, Room No.31317
○ Abstract: In linear algebra problems in the form of b = Ax for given b and A, where the dimension of x is much
greater than that of b, the problem is under-determined, such that there exist infinitely many solutions for
x. However, when a condition of sparsity on x and incoherency on A is imposed, a unique solution of can
still be determined by using a method called compressed sensing (CS). Here, we analyze recent connectome
data revealing the olfactory neural circuits of Drosophila melanogaster and show that a sensing matrix
corresponding to A, defined via the information of synaptic connectivities at the interfaces of projections
neurons (PNs), Kenyon cells (KCs), and mushroom body output neurons (MBONs), fulfills the condition
required for the operation of CS. We find that the CS is effectively at work under the premise of recovering
a sparsely encoded code of odor mixture on PNs. Specifically, the sparse response profile of PNs (x, dim(x)
130) can be reconstructed and correctly classified from a much lower dimensional response profile of MBONs
(b, dim(b) 50) via the sensing matrix (A). We test our framework on hundreds of idealistic and naturalistic
odor expression patterns and demonstrate that the CS can accurately classify the given odor. We show
that when more than five different odors are simultaneously introduced, the performance of signal recovery
degrades significantly.