I completed by Masters in the Paul G. Allen School of Computer Science & Engineering at the University of Washington. I'm advised by Professor Byron Boots in the UW Robot Learning Lab since Spring 2021. My research focuses on Robotic Vision and Deep Learning. I also received my BS degree at the University of Washington. My current research interest lie in end-to-end learning for autonomous driving, imitation learning, and generalizable perception (open set recognition, domain adaptation).
Email: joonl4(at)cs(dot)washington(dot)edu
CV
LinkedIn Profile
Research Blogs
Recently learned about einsum, a flexible operator that can represent a lot of matrix operations. Essentially, einsum conceptually works by writing input and output in terms of axes, which implicitly shows how the output should be calculated.
# setup example matrices
a = np.random()
# permute axis
The & operator assigns reference key to a mapping, while * calls it, and « does a merge of mappings accordingly. Hence for
foo:
a: b
<<:
c: d
e: f
The output comes out as
foo:
a: b
c: d
e: f
While if we are merging them via references, the merging mapping has less priority such that
foo: &foo
a: b
c: d
e: f
fee: &fee
a: q
<<: *foo
far: &far
z: x
<<: *foo
Produces
foo:
a: b
c: d
e: f
fee:
a: q
c: d
e: f
far:
a: b
c: d
e: f
z: x