https://github.com/tomgoldstein/loss-landscape Saddle point https://www.khanacademy.org/math/multivariable-calculus/applications-of- multivariable-derivatives/optimizing-multivariable-functions-videos/v/saddle-points

Conference on Learning Representations (2018). [9] Hamilton, Will, Zhitao Ying, and Jure Leskovec. “Inductive representation learning on large graphs.” Advances in neural information processing systems 30 519-527. 2020. [15] Xu, Da, Chuanwei Ruan, Evren Korpeoglu, Sushant Kumar, and Kannan Achan. “Inductive representation learning on temporal graphs.” International Conference on Learning Representations on Learning Representations. ICLR, 2017. [4] Hamilton, Will, Zhitao Ying, and Jure Leskovec. “Inductive representation learning on large graphs.” Advances in neural information processing systems 30

Nat -> Type) -> -- Property to show (P Z) -> -- Base case ((k : Nat) -> P k -> P (S k)) -> -- Inductive step (x : Nat) -> -- Show for all x P x nat_induction P p_Z p_S Z = p_Z nat_induction P p_Z p_S plus_ind n m = nat_induction (\x => Nat) m -- Base case, plus_ind Z m (\k, k_rec => S k_rec) -- Inductive step plus_ind (S k) m -- where k_rec = plus_ind k m n 奔 奯 奰 奲 奯奶 奥 奴 奨 奡奴 plus n m = plus m n

等长边越多，就越接近圆。这个过程也被称为逼近法（method of exhaustion）。 图2.4.1: 用逼近法求圆的面积 事实上，逼近法就是积分（integral calculus）的起源。2000多年后，微积分的另一支，微分（differential calculus）被发明出来。在微分学最重要的应用是优化问题，即考虑如何把事情做到最好。正如在 2.3.10节中 讨论的那样，这种问题在深度学习中是无处不在的。 6294–6305). 774 Bibliography [McCulloch & Pitts, 1943] McCulloch, W. S., & Pitts, W. (1943). A logical calculus of the ideas immanent in nervous activity. The bulletin of mathematical biophysics, 5(4), 115–133

预览版202112 2.5 参考文献 11 2.5 参考文献 [1] W. S. McCulloch 和 W. Pitts, “A logical calculus of the ideas immanent in nervous activity,” The bulletin of mathematical biophysics, 卷 5, pp.

template, remove all manual dependsOn() calls and source sets created with by creating constructions. To check the list of all default source sets, see the full hierarchy template. If you want and release build types or custom flavors like demo and full . It made them accessible by constructions like val androidDebug by getting { ... } . In the new Android source set layout, those source

hierarchy template, remove all manual dependsOn() calls and source sets created with by creating constructions. To check the list of all default source sets, see the full hierarchy template. If you want and release build types or custom flavors like demo and full . It made them accessible by constructions like val androidDebug by getting { ... } . In the new Android source set layout, those source

Sobel, Prewitt and Kirsch all perform edge detection (in slightly different ways) based on vector calculus and set theory equations. Soften: Slightly blurs the image. Box Sharpen: Increases the contrast Alembic is focused on efficiently storing the computed results of complex procedural geometric constructions. It is very specifically not concerned with storing the complex dependency graph of procedural