LIHKG AI/Machine Learning Book List
我沒有放棄 2018-4-27 22:45:11 唔夠,要學UG year 1 linear algebra + multivariable calculus + statistics

Ads

焦糖布甸 2018-4-27 22:52:23 如果得AS math & stats底其實係咪要由basic algebra學起
:^(
:^(
:^(
我沒有放棄 2018-4-27 22:58:52 起碼要識linear algebra
焦糖布甸 2018-4-27 23:04:07 唔該曬
:^(
:^(
:^(
吾系甘牙吊梨 2018-4-27 23:36:57 Thx 巴打

但我意思係上面咁多個範疇應由邊個睇先(我唔想好似未學加數就學乘數咁)
定係每個範疇都可以平行咁睇 無話邊個特別深/淺
我沒有放棄 2018-4-27 23:43:20 先睇statistical inference同convex optimization,前者係打stat底,後者比較易入口
之後好多其他嘅書都係stat theory,大部分可以同時睇
Probability就要識real analysis (measure theory),可能最難
吾系甘牙吊梨 2018-4-27 23:51:48 THX
:^(
RX-78-2 2018-4-28 00:41:44 淨讀theory唔做programming係乜野嘅學AI
:^(
:^(
:^(


跟本兩邊都要學曬先叫識AI

同埋AI數都好多種,淨係用咩optimization咩情況下用乜野loss function都係一大類

淨係做ANN CNN RNN d數都算淺,玩到capsule net 果堆tensorize咗d operator真係
:^(
:^(
:^(
我沒有放棄 2018-4-28 01:04:49 就係programming網上太多resource,呢個post嘅原意就係想希望大家深入了解machine learning,從theory入手,唔係單單為咗學DL,識用下咩tensorflow run下MLP/CNN/RNN/LSTM

而且呢個post唔係focus落deep learning度,如果你講DL用乜optimization同用乜loss function亦都係一啲好皮毛嘅嘢,反而要明白點解某啲optimization algorithms會快啲converge to local minima/易啲escape saddle points先係重點,了解背後原理嘅永遠係少數人,但呢啲少數人先會行得最遠

另外,Machine learning入面applied math同stat嘅theory十分重要,大部分classical algorithms都係由statistics derive,唔係run下program就會學得識

而家有個現象就係DL嘅熱潮加上scikit-learn/tensorflow/pytorch等libraries嘅出現令到入門門檻低咗,但係反而令到初學者有錯覺係識program就等於識ML,而且deep learning只係machine learning嘅一小部分,唔希望有個錯覺識咗DL等於識machine learning,連簡單嘅linear/logistic regression, SVM, Boosting, Bagging, SVM, PCA等等都唔知點運作
忒修斯之船 2018-4-28 02:03:02 linear regression正常讀過大學都應該識嘅嘢嚟,做ml反而唔識嘅話就真係
:^(
:^(
我沒有放棄 2018-4-28 02:05:17 例子姐,statistical learning其實好博大精深,而且high-dimensional statistics更加同EE/CS關係好密切

Ads

土木人 2018-4-28 06:47:56 我係civil PhD 雖然research同ML冇乜關係,但一直好想學下ML,畢竟係大趨勢!想問下剩係學校Linear Algebra同calculus 夠唔夠讀上面堆書?定要讀其他入門書先?好有心學,唔該巴打先!
高交員 2018-4-28 07:20:46 其實無乜所謂。好似學電腦咁,分好多層面。你嘅興趣係寫game,你都唔會自己去由operation system 學寫上去,只要大概明白OS有乜function,唔洗去研究implementation detail. 當然,當ML初學者program到某一個階段,對啲library有興趣,咁可以深研一下theory.

好多fields 都係咁樣

Applied Science vs Science
Architects vs Civil Engineers
Traders vs Quants
力王 2018-4-28 07:33:24 其實要ML入門的話, 上coursera 已經夠, 部份書上面GE野已經過時
識basic linear algebra, calculus, stat 已經夠做

如果唔滿足只係入門, 而係做research 的話, 就要睇你做邊方面, 如果你想做practical d, 咁可能唔需要太強ge theory 知識

lee 個post 入面D 書都係for ML, AI仲有好多其他野(robotic, planning, scheduling果D) 如果巴打絲打係對其他有興趣就要另外搵書了
plancktime 2018-4-28 08:19:08 香港學背後原理意義唔大
最重要係識用
除非你真係入到Google, Microsoft的Research Team做
plancktime 2018-4-28 08:27:30 如果想commercial application
應該係學呢些
https://academy.microsoft.com/en-us/professional-program/tracks/artificial-intelligence/
RX-78-2 2018-4-28 08:32:36 AI同ML唔係淨係得DL/ANN一途

但係支撐AI發展除左數學,仲需要software engineering, big data處理

你識好深入ML d數,代唔代表好了解AI/ML?其實都只係睇到其中一層

我既應用係computer vision, CNN, kmeans, histogram分析, bayesian, markov random field, stat分類, spatial registration, feature extraction呢類都玩唔少

做AI有分應用層同純theory層,pure theory係low level d, 應用到coding層面abstraction layer會多左,但兩者係相輔相承

另外做AI而家仲有行chipset一途,你話做AI chipset識唔識linear algebra?佢地一定識
計matrix使唔使識Eigen/MKL/BLAS呢類linear algebra專用library?

你鐘意玩既可以由STL build起自己套ML library,但實戰既時候,TF, caffe, pytorch呢d底層都係用BLAS, LAPACK呢堆野架咋
RX-78-2 2018-4-28 08:37:07 information theory同physics 某d數都好似

softmax function = Boltzmann statistics
cross entropy就真頭係thermodynamics果度來
力王 2018-4-28 09:13:16 所以一開頭已經講左係入門
葉斯龍 2018-4-28 13:32:28 我係stat底,岩岩睇完ISLR,應該追邊本好?ISLR 入面d數學好似太少,有時仲難理解,例如:PCA
我沒有放棄 2018-4-28 14:45:29 咁直接睇ESL好過
我沒有放棄 2018-4-28 14:49:18 PhD
:^(
:^(

學埋UG intro Stat就ok
:^(