up

haoxins · Mar 23, 2024 · 8971d3f · 8971d3f
1 parent 2acf5a6
commit 8971d3f
Show file tree

Hide file tree

Showing 5 changed files with 102 additions and 7 deletions.
diff --git a/2024/paper-blog-ml.md b/2024/paper-blog-ml.md
@@ -7,6 +7,10 @@ date: 2023-09-07
 
 ---
 
+
+---
+
+- [Hierarchical Navigable Small Worlds (HNSW)](https://www.pinecone.io/learn/series/faiss/hnsw/)
   - [OasysDB](https://github.com/oasysai/oasysdb)
   - OasysDB is an embeddable, efficient, and
     easy to use vector database. It is designed to be used
@@ -18,10 +22,22 @@ date: 2023-09-07
   - OasysDB implements __HNSW__
     (Hierachical Navigable Small World)
     as its indexing algorithm.
+  - [Benchmarking nearest neighbors](https://github.com/erikbern/ann-benchmarks)
 
 ### 图
 
 
+
+---
+
+- [Graph of Thoughts: Solving Elaborate Problems with Large Language Models](https://arxiv.org/abs/2308.09687)
+  - 2023 (v1), 2024 (v4)
+  - https://github.com/spcl/graph-of-thoughts
+
+```
+
+```
+
 ---
 
 - [A Review on Graph Neural Network Methods in Financial Applications](https://arxiv.org/abs/2111.15367)
@@ -63,6 +79,17 @@ node embedding at each time stamp and then utilize a
 recurrent neural network to aggregate the information.
 ```
 
+```
+How to improve the scalability of GNNs is vital but challenging.
+Computing the Laplacian matrix becomes hard with millions of
+nodes and for a graph of irregular Euclidean space,
+optimizing the algorithm is also difficult.
+Sampling techniques may partially solve the problem with
+the cost of losing structural information.
+Thus, how to maintain the graph structure and improve the
+efficiency of GNN algorithms is worth further exploration.
+```
+
 ---
 
 - [Relational Deep Learning: Graph Representation Learning on Relational Databases](https://arxiv.org/abs/2312.04615)

diff --git a/2024/papers-ml.md b/2024/papers-ml.md
diff --git a/2024/programming.md b/2024/programming.md
@@ -18,6 +18,25 @@ go install golang.org/x/tools/cmd/deadcode@latest
 
 ---
 
+- [Flink Kubernetes Operator 1.8 Release](https://flink.apache.org/2024/03/21/apache-flink-kubernetes-operator-1.8.0-release-announcement/)
+
+```
+Resource savings are nice to have, but the real power of
+Flink Autotuning is the reduced time to production.
+
+With Flink Autoscaling and Flink Autotuning, all users
+need to do is set a max memory size for the TaskManagers,
+just like they would normally configure TaskManager memory.
+Flink Autotuning then automatically adjusts the various
+memory pools and brings down the total container memory size.
+It does that by observing the actual max memory usage on
+the TaskMangers or by calculating the exact number of
+network buffers required for the job topology.
+The adjustments are made together with Flink Autoscaling,
+so there is no extra downtime involved.
+```
+
+> 很实用的理念, 实际效果有待检验!
 
 - [The magic of Rust's type system](https://www.youtube.com/watch?v=NDIU1GSBrVI&t=11s)
 

diff --git a/2024/quantum-computation-1.md b/2024/quantum-computation-1.md
@@ -1626,6 +1626,47 @@ RSA 的加密信息仍需要比当前量子计算机的规模扩大五个数量
     以区别于密度算子
     $$ ρ $$.
 
+---
+
+- 最后, 设想量子状态以概率
+  $$ p_i $$
+  处于状态
+  $$ ρ_i $$.
+  不难发现系统可以用密度矩阵
+  $$ \sum_{i} p_i ρ_i $$
+  来描述. 关于这一点的证明如下:
+  - 设
+    $$ ρ_i $$
+    来自某个纯态的系综
+    $$ \{ p_{ij}, \mid ψ_{ij} \rangle \} $$
+    (注意 `i` 是固定的), 于是处于状态
+    $$ \mid ψ_{ij} \rangle $$
+    的概率是
+    $$ p_i p_{ij} $$,
+    因此系统的密度矩阵是
+  - $$
+      \begin{align}
+        ρ & =
+            \sum_{i, j} p_i p_{ij}
+            \mid ψ_{ij} \rangle \langle ψ_{ij} \mid \\
+          & =
+            \sum_{i} p_i ρ_i
+      \end{align}
+    $$
+  - 其中用了定义
+    $$
+      ρ_i = \sum_{j} p_{ij}
+      \mid ψ_{ij} \rangle \langle ψ_{ij} \mid
+    $$,
+    称
+    $$ ρ $$
+    为具有概率
+    $$ p_i $$
+    的状态
+    $$ ρ_i $$
+    的混合. 这个混合的概念在如量子噪声问题的分析中反复出现,
+    其中噪声的影响为我们对量子状态的认识引入了不确定性.
+
 ### 施密特分解与纯化
 
 ### EPR 和贝尔不等式

diff --git a/2024/reading.md b/2024/reading.md
@@ -7,14 +7,27 @@ date: 2023-07-17
 
 - [拓扑与物理](https://book.douban.com/subject/35388292/)
 
-- [惰者集](https://book.douban.com/subject/27603987/)
-
 - [黑体辐射](https://book.douban.com/subject/36701663/)
   - __阅读之前__: 终于有一本细分点切入, 一以贯之的书了~
     市面上泛泛而谈的科普书太多了, 不是量子就是相对论,
     照葫芦画瓢, 没啥特色~
   - 同时支持一下国人作品!
 
+- [惰者集](https://book.douban.com/subject/27603987/)
+  - 感知!
+  - 行文轻松, 语言平实. 很适合闲来碎片化阅读!
+
+```
+也就是说, 数学运算支配了作为量子力学对象的物理现象.
+这种数学运算与物理现象的关系,
+并非是通过解析叠加的物理意义而将其用数学公式表现出来,
+而是将 "波函数的线性组合可以描述状态的叠加" 视为公理,
+然后依据数学运算来确定叠加的意义.
+正如费曼所言, 除了数学之外, 没有其他方法能说明态叠加原理了.
+我们只能认为量子力学基于数学的无穷魔法,
+因此我认为物理现象的背后存在着固有的数学现象.
+```
+
 - [图解 TCP/IP (第 6 版)](https://book.douban.com/subject/36739612/)
   - 有些年头没看网络相关的东西了~
   - 废话不少~