Propose reading group papers on slack. Formatted with style.
git clone https://github.com/andreasgrv/emojivote
cd emojivote
python3 -m venv .env
source .env/bin/activate
pip install -r requirements.txt
pip install .- Run the installed command, passing a list of links to the papers you want on arxiv.
emojivote-slack --paper-links https://arxiv.org/abs/2102.11582 https://arxiv.org/abs/math/9701207This should output (emoji selection is random):
:ocean: *Deterministic Neural Networks with Inductive Biases Capture Epistemic and Aleatoric Uncertainty*
_Mukhoti, Jishnu, Kirsch, Andreas, van Amersfoort, Joost, Torr, Philip H. S., Gal, Yarin_
https://arxiv.org/pdf/2102.11582.pdf
We show that a single softmax neural net with minimal changes can beat the
uncertainty predictions of Deep Ensembles and other more complex
single-forward-pass uncertainty approaches. Standard softmax neural nets suffer
from feature collapse and extrapolate arbitrarily for OoD points. This results
in arbitrary softmax entropies for OoD points which can have high entropy, low,
or anything in between, thus cannot capture epistemic uncertainty reliably. We
prove that this failure lies at the core of "why" Deep Ensemble Uncertainty
works well. Instead of using softmax entropy, we show that with appropriate
inductive biases softmax neural nets trained with maximum likelihood reliably
capture epistemic uncertainty through their feature-space density. This density
is obtained using simple Gaussian Discriminant Analysis, but it cannot
represent aleatoric uncertainty reliably. We show that it is necessary to
combine feature-space density with softmax entropy to disentangle uncertainties
well. We evaluate the epistemic uncertainty quality on active learning and OoD
detection, achieving SOTA ~98 AUROC on CIFAR-10 vs SVHN without fine-tuning on
OoD data.
:snowflake: *Piles of Cubes, Monotone Path Polytopes and Hyperplane Arrangements*
_Athanasiadis, Christos A._
https://arxiv.org/pdf/math/9701207.pdf
Monotone path polytopes arise as a special case of the construction of fiber
polytopes, introduced by Billera and Sturmfels. A simple example is provided by
the permutahedron, which is a monotone path polytope of the standard unit cube.
The permutahedron is the zonotope polar to the braid arrangement. We show how
the zonotopes polar to the cones of certain deformations of the braid
arrangement can be realized as monotone path polytopes. The construction is an
extension of that of the permutahedron and yields interesting connections
between enumerative combinatorics of hyperplane arrangements and geometry of
monotone path polytopes.
- Paste the above into Slack and use
Ctrl+Shift+Fto Format them message.
Idea inspired and slack format created by Kate McCurdy