Google、軟銀都陣亡過!盤點 AI 專案失敗的 4 大原因
Posted on2021/02/04
若水AI Blog
【我們為什麼挑選這篇文章】為了適應未知多變的世界,許多企業搶做「數位轉型」,從公司營運的各層面如客戶體驗、商業模式、企業文化到作業流程等,透過科技的導入來提升效率與效能;而對製造業企業而言,原料採購、物流管理、庫存調配、生產、行銷等環節則是企業主進行數位轉型會優先考量的面向。
在這之中,AI 的運用扮演很關鍵的角色,如何將 AI 應用到上述各層面並實際執行,是許多企業面臨的挑戰,有哪些要點是執行 AI 專案時需特別留意的?(責任編輯:賴佩萱)
作者:若水 AI 資料資料處理部負責人 簡季婕
2020 年,突如其來的新冠肺炎疫情(Covid-19)改變了許多產業的命運,同時加速推促 AI 落地的速度,AI 人工智慧的應用將成為企業的新日常。
若水 AI 資料服務團隊本著為臺灣 AI 應用落地盡份心力的初衷,順著這波改變,推出全新系列內容:與機器學習(ML : Machine Learning)、AIOps 智慧運維(Artificial Intelligence for IT Operations)有關的實用文,分享各界專家在每一天如何持續營運、優化 AI 架構以及資料處理的基本功。
【若水導讀】AI 專案順利通關的三個絕招:
1. AI 資料來源要多元,避免學習偏誤
2. 標註前,請先建立客觀的 AI 資料標註(Data Annotation)原則
3. 讓 AI 人工智慧成為組織的共同語言,會更容易成功
企業都想做 AI,但實際上沒那麼簡單
根據《臺灣人工智慧學校 AI Academy Taiwan》2019 年針對臺灣各大產業 1,095 位業界校友的調查統計,成功導入 AI 人工智慧的臺灣企業僅占 20%。放眼國際,許多全球知名企業的 AI 專案也慘遭滑鐵盧:
Google 在泰國落地測試智慧醫療失敗,拖慢醫療流程;美國杜克大學發佈的 PULSE 演算法誤將歐巴馬的頭像還原為白人,引發種族歧視爭議。
在日本,軟銀(Softbank)社長孫正義原本打算以 AI 機器人取代銷售人員,沒想到 AI 機器人無法應付實際場域的複雜性,計畫負責人只好承認失敗:「我們把機器學習(Machine Learning)想得太簡單了」。
AI 專案難實際執行,問題出在哪?
若水經手過臺灣、日本超過 200 個的 AI 資料處理專案,從橫跨各大產業領域的專案經驗,整理出企業 AI 之所以無法順利落地的四大原因。
1. AI 模型訓練過程中沒有加入實際場域的資料
無論是剛導入 AI 而產生資料處理需求的新手企業,還是已有 AI 專案經驗、為了 retrain 模型再度找上若水的老手企業,都曾經在同一個地方卡關:AI 資料標註品質有做到位元,但 AI 模型卻無法應用落地 。
為什麼?
原因在於,客戶並未以「實際場景」的資料來進行 AI 模型訓練。
現在市面上有許多開放資料集(Open Dataset)或是免費的商用網路圖片,企業通常會優先使用這些免費資源進行 AI 資料標註(Data Annotation)讓機器學習,但是放到實際場域測試後,經常發現 AI 模型成效不佳,無法適用於實際場景,最終還是需要回過頭再進行第二次模型訓練(Model Training)。
因此 在 AI 專案開始前,建議企業首先需要在內部建立資料資料流(Data Pipeline),而在收集資料時,不只使用開放資料集(Open Dataset),也須確保有使用符合實際應用場景的資料來訓練 AI 模型,全盤考量資料類型、角度等多元性,避免機器學習偏誤 。
2. AI 資料標註原則定義不夠客觀
與企業工程師對接 AI 資料處理需求時,當我們詢問這批人臉辨識(Face Recognition)的 AI 資料標註的原則是什麼,常常會接到諸如此類的回答:「頭太小的話,就不要標註數據」。
一般人的邏輯覺得很合理的事情,對於機器學習(Machine Learning)來說卻是一大挑戰。 機器學習需要知道的是趨近「絕對客觀」的原則 ,例如,所謂的頭太大、太小,換算成具體數值會是幾乘幾大小的 pixel?如果圖片背景融色或模糊,也需要標註起來嗎?
一旦 AI 資料標註原則不夠客觀,AI 模型很容易隨著人的「主觀認定」來學習,當專案換了一位工程師,機器學習出來的效果可能也會跟著變 。在我們的經驗,原則的訂定最好透過「對話」,藉由反覆詰問,才能加快釐清目標。有了歸納、定義出客觀的 AI 資料標註原則。就會加快模型學習(Model Learning)成效。
為了清楚定義圖片融色或模糊的問題,我們採用國際照明委員會(International Commission on Illumination)訂定的 Delta E 標準,和影像(圖像)品質評估標準 BRISQUE,和客戶確認彼此認知是否一致。
根據國際標準,人的肉眼能分辨得出來的色差,至少會在 Delta E 值 2 以上。所以,當一張影像測出來 Delta E 值小於 2,就表示這張圖的融色程度太高,無法標註。
假如客戶希望「太模糊的圖片不要標註」,團隊也會根據 BRISQUE(影像品質評估標準)的標準,輸出不同模糊指數的圖片,請客戶確認所謂的模糊,具體來說是 70% 還是 80%。
3. AI 模型訓練(Model Training)沒有循序漸進
以肢體行為辨識(Posture Estimation)為例,Coco Dataset 從一開始只辨識人體 7 大主要關鍵點(Key Point),後來逐步發展成 25 點,甚至快 40 點,有些客戶會希望若水 AI 團隊可以一次就標註 40 個關鍵點,直接拿去機器學習(Machine Learning)。
說起來,機器學習和教小孩很像,一下子給太多的特徵點(Feature Points)反而會「揠苗助長」,導致 AI 模型學到最後分不清楚自己到底在學習什麼。我們也遇過有些客戶,一開始想用難度較高的 Segmentation 方式讓模型學習人的行為,但是人的行為百百種、語意切割(Segmentation)的變異度也高,就比較難學得好。
當這些客戶再回頭來找若水,通常會比較循序漸進,從小地方開始逐步改進 AI 模型。
4. 缺乏管理層的理解與支持
AI 熱潮讓許多企業趨之若鶩,然而 AI 要能夠順利落地,除了上述三項實務建議,企業管理層對於 AI 的認知和支持更是一大關鍵。
許多臺灣企業的 AI 數位轉型主導者,可能是傳統公司裡面有豐富資歷的 CTO 技術長或管理階層,對於 AI 人工智慧這個全新領域的概念,比較缺乏深度的理解,也沒有類似 AI 模型訓練和測試的相關經驗,從上述 4 個原因去追尋難以落地的根源,或許能有所助益。
資料來源:https://buzzorange.com/techorange/2021/02/04/ai-project-difficulties/?fbclid=IwAR04ZC1-1MquyCObEI5HIfTKtV-OkcfxL_R8vRin4YgQMl8cnhS_6aM59vU
同時也有735部Youtube影片,追蹤數超過6萬的網紅Herman Yeung,也在其Youtube影片中提到,M1, M2 Free Note download 免費筆記下載 : https://hermanutube.blogspot.hk/2016/01/youtube-pdf.html Past Paper (香港公共圖書館): https://mmis.hkpl.gov.hk/web/guest...
「point estimation」的推薦目錄:
- 關於point estimation 在 台灣物聯網實驗室 IOT Labs Facebook 的精選貼文
- 關於point estimation 在 元毓 Facebook 的最佳貼文
- 關於point estimation 在 Herman Yeung Facebook 的最讚貼文
- 關於point estimation 在 Herman Yeung Youtube 的最佳解答
- 關於point estimation 在 Herman Yeung Youtube 的最佳解答
- 關於point estimation 在 Herman Yeung Youtube 的最佳解答
- 關於point estimation 在 Chapter 10 Point Estimation | Introduction to Statistical Thinking 的評價
- 關於point estimation 在 Point estimation, interval estimation, density estimation? 的評價
point estimation 在 元毓 Facebook 的最佳貼文
根據計算,100萬人遊行隊伍要從維多利亞公園排到廣東;200萬人遊行則要排到泰國。
順道一提香港15~30歲人口約莫100出頭萬人。以照片人群幾乎都是此年齡帶來看,兩個數字都是明顯誇大太多了。
另一個可以參考的是1969年的Woodstock Music & Art Fair,幾天內湧進40萬人次,照片看起來也是滿山滿谷的人。(http://sites.psu.edu/…/upl…/sites/851/2013/01/Woodstock3.jpg)
當年40萬人次引發驚人的大塞車,幾乎花十幾個小時才逐漸清場。
而香港遊行清場速度明顯快得多。
順道一提,因此運動而認定「你的父母不愛你」的白痴論述也如同文化大革命時的「爹親娘親不如毛主席親」般開始出現:
https://www.facebook.com/SaluteToHKPolice/videos/350606498983830/UzpfSTUyNzM2NjA3MzoxMDE1NjMyMTM4NjY3MTA3NA/
EVERY MAJOR NEWS outlet in the world is reporting that two million people, well over a quarter of our population, joined a single protest.
.
It’s an astonishing thought that filled an enthusiastic old marcher like me with pride. Unfortunately, it’s almost certainly not true.
.
A march of two million people would fill a street that was 58 kilometers long, starting at Victoria Park in Hong Kong and ending in Tanglangshan Country Park in Guangdong, according to one standard crowd estimation technique.
.
If the two million of us stood in a queue, we’d stretch 914 kilometers (568 miles), from Victoria Park to Thailand. Even if all of us marched in a regiment 25 people abreast, our troop would stretch towards the Chinese border.
.
Yes, there was a very large number of us there. But getting key facts wrong helps nobody. Indeed, it could hurt the protesters more than anyone.
.
For math geeks only, here’s a discussion of the actual numbers that I hope will interest you whatever your political views.
.
.
DO NUMBERS MATTER?
.
People have repeatedly asked me to find out “the real number” of people at the recent mass rallies in Hong Kong.
.
I declined for an obvious reason: There was a huge number of us. What does it matter whether it was hundreds of thousands or a million? That’s not important.
.
But my critics pointed out that the word “million” is right at the top of almost every report about the marches. Clearly it IS important.
.
.
FIRST, THE SCIENCE
.
In the west, drone photography is analyzed to estimate crowd sizes.
.
This reporter apologizes for not having found a comprehensive database of drone images of the Hong Kong protests.
.
But we can still use related methods, such as density checks, crowd-flow data and impact assessments. Universities which have gathered Hong Kong protest march data using scientific methods include Hong Kong Polytechnic University, Hong Kong University of Science and Technology, University of Hong Kong, and Hong Kong Baptist University.
.
.
DENSITY CHECKS
.
Figures gathered in the past by Hong Kong Polytechnic specialists using satellite photo analysis found a density level of one square meter per marcher. Modern analysis suggests this remains roughly accurate.
.
I know from experience that Hong Kong marches feature long periods of normal spacing (one square meter or one and half per person, walking) and shorter periods of tight spacing (half a square meter or less per person, mostly standing).
.
.
JOINERS AND SPEED
.
We need to include people who join halfway. In the past, a Hong Kong University analysis using visual counting methods cross-referenced with one-on-one interviews indicated that estimates should be boosted by 12% to accurately reflect late joiners. These days, we’re much more generous in estimating joiners.
.
As for speed, a Hong Kong Baptist University survey once found a passing rate of 4,000 marchers every ten minutes.
.
Videos of the recent rallies indicates that joiner numbers and stop-start progress were highly erratic and difficult to calculate with any degree of certainty.
.
.
DISTANCE MULTIPLIED BY DENSITY
.
But scientists have other tools. We know the walking distance between Victoria Park and Tamar Park is 2.9 kilometers. Although there was overspill, the bulk of the marchers went along Hennessy Road in Wan Chai, which is about 25 meters (or 82 feet) wide, and similar connected roads, some wider, some narrower.
.
Steve Doig, a specialist in crowd analysis approached by the Columbia Journalism Review (CJR), analyzed an image of Hong Kong marchers to find a density level of 7,000 people in a 210-meter space. Although he emphasizes that crowd estimates are never an exact science, that figure means one million Hong Kong marchers would need a street 18.6 miles long – which is 29 kilometers.
.
Extrapolating these figures for the June 16 claim of two million marchers, you’d need a street 58 kilometers long.
.
Could this problem be explained away by the turnover rate of Hong Kong marchers, which likely allowed the main (three kilometer) route to be filled more than once?
.
The answer is yes, to some extent. But the crowd would have to be moving very fast to refill the space a great many times over in a single afternoon and evening. It wasn’t. While I can walk the distance from Victoria Park to Tamar in 41 minutes on a quiet holiday afternoon, doing the same thing during a march takes many hours.
.
More believable: There was a huge number of us, but not a million, and certainly not two million.
.
.
IMPACT MEASUREMENTS
.
A second, parallel way of analyzing the size of the crowd is to seek evidence of the effects of the marchers’ absence from their normal roles in society.
.
If we extract two million people out of a population of 7.4 million, many basic services would be severely affected while many others would grind to a complete halt.
.
Manpower-intensive sectors of society, such as transport, would be badly affected by mass absenteeism. Industries which do their main business on the weekends, such as retail, restaurants, hotels, tourism, coffee shops and so on would be hard hit. Round-the-clock operations such as hospitals and emergency services would be severely troubled, as would under-the-radar jobs such as infrastructure and utility maintenance.
.
There seems to be no evidence that any of that happened in Hong Kong.
.
.
HOW DID WE GET INTO THIS MESS?
.
To understand that, a bit of historical context is necessary.
.
In 2003, a very large number of us walked from Victoria Park to Central. The next day, newspapers gave several estimates of crowd size.
.
The differences were small. Academics said it was 350,000 plus. The police counted 466,000. The organizers, a group called the Civil Rights Front, rounded it up to 500,000.
.
No controversy there. But there was trouble ahead.
.
.
THINGS FALL APART
.
At a repeat march the following year, it was obvious to all of us that our numbers were far lower that the previous year. The people counting agreed: the academics said 194,000 and the police said 200,000.
.
But the Civil Rights Front insisted that there were MORE than the previous year’s march: 530,000 people.
.
The organizers lost credibility even with us, their own supporters. To this day, we all quote the 2003 figure as the high point of that period, ignoring their 2004 invention.
.
.
THE TRUTH COUNTS
.
The organizers had embarrassed the marchers. The following year several organizations decided to serve us better, with detailed, scientific counts.
.
After the 2005 march, the academics said the headcount was between 60,000 and 80,000 and the police said 63,000. Separate accounts by other independent groups agreed that it was below 100,000.
.
But the organizers? The Civil Rights Front came out with the awkward claim that it was a quarter of a million. Ouch. (This data is easily confirmed from multiple sources in newspaper archives.)
.
.
AN UNEXPECTED TWIST
.
But then came a twist. Some in the Western media chose to present ONLY the organizer’s “outlier” claim.
.
“Dressed in black and chanting ‘one man, one vote’, a quarter of a million people marched through Hong Kong yesterday,” said the Times of London in 2005.
.
“A quarter of a million protesters marched through Hong Kong yesterday to demand full democracy from their rulers in Beijing,” reported the UK Independent.
.
It became obvious that international media outlets were committed to emphasizing whichever claim made the Hong Kong government (and by extension, China) look as bad as possible. Accuracy was nowhere in the equation.
.
.
STRATEGICALLY CHOSEN
.
At universities in Hong Kong, there were passionate discussions about the apparent decision to pump up the numbers as a strategy, with the international media in mind. Activists saw two likely positive outcomes.
.
First, anyone who actually wanted the truth would choose a middle point as the “real” number: thus it was worth making the organizers’ number as high as possible. (The police could be presented as corrupt puppets of Beijing.)
.
Second, international reporters always favored the largest number, since it implicitly criticized China. Once the inflated figure was established in the Western media, it would become the generally accepted figure in all publications.
.
Both of the activists’ predictions turned out to be bang on target. In the following years, headcounts by social scientists and police were close or even impressively confirmed the other—but were ignored by the agenda-driven international media, who usually printed only the organizers’ claims.
.
.
SKIP THIS SECTION
.
Skip this section unless you want additional examples to reinforce the point.
.
In 2011, researchers and police said that between 63,000 and 95,000 of us marched. Our delightfully imaginative organizers multiplied by four to claim there were 400,000 of us.
.
In 2012, researchers and police produced headcounts similar to the previous year: between 66,000 and 97,000. But the organizers claimed that it was 430,000. (These data can also be easily confirmed in any newspaper archive.)
.
.
SKIP THIS SECTION TOO
.
Unless you’re interested in the police angle. Why are police figures seen as lower than others? On reviewing data, two points emerge.
.
First, police estimates rise and fall with those of independent researchers, suggesting that they function correctly: they are not invented. Many are slightly lower, but some match closely and others are slightly higher. This suggests that the police simply have a different counting method.
.
Second, police sources explain that live estimates of attendance are used for “effective deployment” of staff. The number of police assigned to work on the scene is a direct reflection of the number of marchers counted. Thus officers have strong motivation to avoid deliberately under-estimating numbers.
.
.
RECENT MASS RALLIES
.
Now back to the present: this hot, uncomfortable summer.
.
Academics put the 2019 June 9 rally at 199,500, and police at 240,000. Some people said the numbers should be raised or even doubled to reflect late joiners or people walking on parallel roads. Taking the most generous view, this gave us total estimates of 400,000 to 480,000.
.
But the organizers, God bless them, claimed that 1.03 million marched: this was four times the researchers’ conservative view and more than double the generous view.
.
The addition of the “.03m” caused a bit of mirth among social scientists. Even an academic writing in the rabidly pro-activist Hong Kong Free Press struggled to accept it. “Undoubtedly, the anti-amendment group added the extra .03 onto the exact one million figure in order to give their estimate a veneer of accuracy,” wrote Paul Stapleton.
.
.
MIND-BOGGLING ESTIMATE
.
But the vast majority of international media and social media printed ONLY the organizers’ eyebrow-raising claim of a million plus—and their version soon fed back into the system and because the “accepted” number. (Some mentioned other estimates in early reports and then dropped them.)
.
The same process was repeated for the following Sunday, June 16, when the organizers’ frankly unbelievable claim of “about two million” was taken as gospel in the majority of international media.
.
“Two million people in Hong Kong protest China's growing influence,” reported Fox News.
.
“A record two million people – over a quarter of the city’s population” joined the protest, said the Guardian this morning.
.
“Hong Kong leader apologizes as TWO MILLION take to the streets,” said the Sun newspaper in the UK.
.
Friends, colleagues, fellow journalists—what happened to fact-checking? What happened to healthy skepticism? What happened to attempts at balance?
.
.
CONCLUSIONS?
.
I offer none. I prefer that you do your own research and draw your own conclusions. This is just a rough overview of the scientific and historical data by a single old-school citizen-journalist working in a university coffee shop.
.
I may well have made errors on individual data points, although the overall message, I hope, is clear.
.
Hong Kong people like to march.
.
We deserve better data.
.
We need better journalism. Easily debunked claims like “more than a quarter of the population hit the streets” help nobody.
.
International media, your hostile agendas are showing. Raise your game.
.
Organizers, stop working against the scientists and start working with them.
.
Hong Kong people value truth.
.
We’re not stupid. (And we’re not scared of math!)
point estimation 在 Herman Yeung Facebook 的最讚貼文
M1, M2 的同學留意︰
因為有太多同學問翻部分我地未拍的 past paper solution
所以我地今日再補拍翻之前 miss 左的 past paper solution 出黎
暫時有呢 8 條,我地會繼續拍
可能會拍其他缺的題目
希望大家可以更有效溫書,加油﹗
Herman Yeung - DSE Maths (M2) Sample Paper/Q9 (3D vector)
https://youtu.be/EcxKS_w7ivs
Herman Yeung - DSE Maths (M2) Practice Paper/Q2 (System of linear equation)
https://youtu.be/xE0tmYcctw0
Herman Yeung - DSE Maths (M2) Practice Paper/Q3 (Out-syllabus)
https://youtu.be/jZeO9TSrwDo
Herman Yeung - DSE Maths (M1) PP Sample Paper Q06 (Differentiation)
https://youtu.be/5OG9JE49-Ro
Herman Yeung - DSE Maths (M1) PP Sample Paper Q07 (Point & Interval Estimation)
https://youtu.be/36tWbTKW4jQ
Herman Yeung - DSE Maths (M1) PP Sample Paper Q13 (4 distributions)
https://youtu.be/L0SIxlJTJvo
Herman Yeung - DSE Maths (M1) PP Practice Paper Q05 (Application of integration)
https://youtu.be/tyNMx2evO9s
Herman Yeung - DSE Maths (M1) PP Practice Paper Q07 (Bayes' Theorem)
https://youtu.be/GfT5OOoQuI4
------------------------------------
HKDSE Mathematics 數學天書 訂購表格及方法︰
http://goo.gl/forms/NgqVAfMVB9
(**訂購天書,一般今日完成程序、明天寄、後天收到 (假期除外))
YouTube 網上教學平台 : http://www.youtube.com/HermanYeung
Herman Yeung Blogger : https://goo.gl/SBmVOO
Herman Yeung Instagram : https://www.instagram.com/hermanyeung_hy
point estimation 在 Herman Yeung Youtube 的最佳解答
M1, M2 Free Note download 免費筆記下載 : https://hermanutube.blogspot.hk/2016/01/youtube-pdf.html
Past Paper (香港公共圖書館): https://mmis.hkpl.gov.hk/web/guest/hkcee-and-hkale-papers-collection
------------------------------------------------------------------------------
M1 所有 videos 的 Playlist 可看: https://goo.gl/l3gAUQ
分類的 Playlist 可看:
https://goo.gl/rlbmEB ……… M1 (Binomial Theorem 二項式定理)
https://goo.gl/FZotov ……… M1 (Exponential & Log. functions 指數對數函數)
https://goo.gl/bx9Gp9 ……… M1 (Differentiation & its application 微分及其應用)
https://goo.gl/8qEBQ0 ……… M1 (Integration & its application 積分及其應用)
https://goo.gl/LEyZVD ……… M1 (Bayes' Theorem貝葉斯定理)
https://goo.gl/BAXGWk ……… M1 (Normal Distribution 正態分佈)
https://goo.gl/sEgQx9 ……… M1 (4 Distributions 四大分佈)
https://goo.gl/PAuvHb ……… M1 (Point & Interval Estimation 點與間距估計)
https://goo.gl/IUCu4a ……… M1 (Tips Class & Last Hour)
------------------------------------------------------------------------------
HKDSE Mathematics 數學天書 訂購表格及方法︰ http://goo.gl/forms/NgqVAfMVB9
課程簡介︰ https://youtu.be/Rgm7yUVG9cY
------------------------------------------------------------------------------
HKDSE 數學 Core 各天書 的內容︰ https://www.facebook.com/hy.publishing/photos/a.312736375489291.68655.198063650289898/933817946714461/?type=3&theater
HKDSE 數學 Core 特別快車班
28堂 (共7本天書) 完成整個 HKDSE 數學 Core
(中一至中六) 要考的所有課題,
適合任何考 HKDSE 的同學上課 (中四至中六都合適)
(p.s. Herman Yeung 所有天書,中英對照)
------------------------------------------------------------------------------
Please subscribe 請訂閱︰
https://www.youtube.com/hermanyeung?sub_confirmation=1
------------------------------------------------------------------------------
Blogger︰ https://hermanutube.blogspot.hk/2016/02/herman-yeung-main-menu.html
Facebook︰ https://www.facebook.com/hy.page
YouTube︰ https://www.youtube.com/HermanYeung
Instagram︰ https://www.instagram.com/hermanyeung_hy
------------------------------------------------------------------------------

point estimation 在 Herman Yeung Youtube 的最佳解答
M1, M2 Free Note download 免費筆記下載 : https://hermanutube.blogspot.hk/2016/01/youtube-pdf.html
Past Paper (香港公共圖書館): https://mmis.hkpl.gov.hk/web/guest/hkcee-and-hkale-papers-collection
------------------------------------------------------------------------------
M1 所有 videos 的 Playlist 可看: https://goo.gl/l3gAUQ
分類的 Playlist 可看:
https://goo.gl/rlbmEB ……… M1 (Binomial Theorem 二項式定理)
https://goo.gl/FZotov ……… M1 (Exponential & Log. functions 指數對數函數)
https://goo.gl/bx9Gp9 ……… M1 (Differentiation & its application 微分及其應用)
https://goo.gl/8qEBQ0 ……… M1 (Integration & its application 積分及其應用)
https://goo.gl/LEyZVD ……… M1 (Bayes' Theorem貝葉斯定理)
https://goo.gl/BAXGWk ……… M1 (Normal Distribution 正態分佈)
https://goo.gl/sEgQx9 ……… M1 (4 Distributions 四大分佈)
https://goo.gl/PAuvHb ……… M1 (Point & Interval Estimation 點與間距估計)
https://goo.gl/IUCu4a ……… M1 (Tips Class & Last Hour)
------------------------------------------------------------------------------
HKDSE Mathematics 數學天書 訂購表格及方法︰ http://goo.gl/forms/NgqVAfMVB9
課程簡介︰ https://youtu.be/Rgm7yUVG9cY
------------------------------------------------------------------------------
HKDSE 數學 Core 各天書 的內容︰ https://www.facebook.com/hy.publishing/photos/a.312736375489291.68655.198063650289898/933817946714461/?type=3&theater
HKDSE 數學 Core 特別快車班
28堂 (共7本天書) 完成整個 HKDSE 數學 Core
(中一至中六) 要考的所有課題,
適合任何考 HKDSE 的同學上課 (中四至中六都合適)
(p.s. Herman Yeung 所有天書,中英對照)
------------------------------------------------------------------------------
Please subscribe 請訂閱︰
https://www.youtube.com/hermanyeung?sub_confirmation=1
------------------------------------------------------------------------------
Blogger︰ https://hermanutube.blogspot.hk/2016/02/herman-yeung-main-menu.html
Facebook︰ https://www.facebook.com/hy.page
YouTube︰ https://www.youtube.com/HermanYeung
Instagram︰ https://www.instagram.com/hermanyeung_hy
------------------------------------------------------------------------------

point estimation 在 Herman Yeung Youtube 的最佳解答
M1, M2 Free Note download 免費筆記下載 : https://hermanutube.blogspot.hk/2016/01/youtube-pdf.html
Past Paper (香港公共圖書館): https://mmis.hkpl.gov.hk/web/guest/hkcee-and-hkale-papers-collection
------------------------------------------------------------------------------
M1 所有 videos 的 Playlist 可看: https://goo.gl/l3gAUQ
分類的 Playlist 可看:
https://goo.gl/rlbmEB ……… M1 (Binomial Theorem 二項式定理)
https://goo.gl/FZotov ……… M1 (Exponential & Log. functions 指數對數函數)
https://goo.gl/bx9Gp9 ……… M1 (Differentiation & its application 微分及其應用)
https://goo.gl/8qEBQ0 ……… M1 (Integration & its application 積分及其應用)
https://goo.gl/LEyZVD ……… M1 (Bayes' Theorem貝葉斯定理)
https://goo.gl/BAXGWk ……… M1 (Normal Distribution 正態分佈)
https://goo.gl/sEgQx9 ……… M1 (4 Distributions 四大分佈)
https://goo.gl/PAuvHb ……… M1 (Point & Interval Estimation 點與間距估計)
https://goo.gl/IUCu4a ……… M1 (Tips Class & Last Hour)
------------------------------------------------------------------------------
HKDSE Mathematics 數學天書 訂購表格及方法︰ http://goo.gl/forms/NgqVAfMVB9
課程簡介︰ https://youtu.be/Rgm7yUVG9cY
------------------------------------------------------------------------------
HKDSE 數學 Core 各天書 的內容︰ https://www.facebook.com/hy.publishing/photos/a.312736375489291.68655.198063650289898/933817946714461/?type=3&theater
HKDSE 數學 Core 特別快車班
28堂 (共7本天書) 完成整個 HKDSE 數學 Core
(中一至中六) 要考的所有課題,
適合任何考 HKDSE 的同學上課 (中四至中六都合適)
(p.s. Herman Yeung 所有天書,中英對照)
------------------------------------------------------------------------------
Please subscribe 請訂閱︰
https://www.youtube.com/hermanyeung?sub_confirmation=1
------------------------------------------------------------------------------
Blogger︰ https://hermanutube.blogspot.hk/2016/02/herman-yeung-main-menu.html
Facebook︰ https://www.facebook.com/hy.page
YouTube︰ https://www.youtube.com/HermanYeung
Instagram︰ https://www.instagram.com/hermanyeung_hy
------------------------------------------------------------------------------

point estimation 在 Point estimation, interval estimation, density estimation? 的推薦與評價
Frequentist statistics textbooks typically consider point and interval estimation but not density estimation of a parameter. ... <看更多>
point estimation 在 Chapter 10 Point Estimation | Introduction to Statistical Thinking 的推薦與評價
Estimate parameters from data and assess the performance of the estimation procedure. 10.2 Estimating Parameters. Statistic is the science of data analysis. The ... ... <看更多>