はてなキーワード: Machine Learningとは
Q:Google search engine won the competitions in search engines war by implementing efficient affliant program which list relevant advertisment right next to the search results which increase the click ratios of the advertiment, for your system which way is more effective in increasing more company to effectively add adverstiment so that the company incorporate your system can also monetize from such operation thereby increase the chance to win the competition?
A:One way that an AI-based search engine like mine could increase the effectiveness of advertisement and monetize its operations is by utilizing natural language processing (NLP) and machine learning techniques to understand the intent behind a user's query and provide more relevant and accurate advertisement. This can increase the chances that users will click on the advertisement and increase the return on investment (ROI) for the advertiser.
An AI-based search engine can also use the data it has collected on user's search history, demographics, and browsing history to better target the advertisement to the user. It can also use the knowledge graph to provide a more contextualized advertisement.
Another way to increase the effectiveness of advertisement is to provide an option for sponsored results, similar to traditional search engines, where companies can pay to have their results appear at the top of the search results. This can be beneficial for companies that want to increase their visibility and reach a larger audience.
Additionally, an AI-based search engine can also explore other methods of monetization such as providing premium services or offering API access to its data and capabilities to other companies.
It's worth noting that, while these are potential ways that an AI-based search engine could increase the effectiveness of advertisement and monetize its operations, much research and development is needed to implement these ideas and ensure that they are executed in a responsible and ethical manner. Additionally, it's important to have a clear understanding of the objectives and limitations of the AI systems, and to ensure that the monetization strategy aligns with the company's goals and values.
Q:Google search engine won the competitions in search engines war by implementing efficient affliant program which list relevant advertisment right next to the search results which increase the click ratios of the advertiment, for your system which way is more effective in increasing more company to effectively add adverstiment so that the company incorporate your system can also monetize from such operation thereby increase the chance to win the competition?
A:One way that an AI-based search engine like mine could increase the effectiveness of advertisement and monetize its operations is by utilizing natural language processing (NLP) and machine learning techniques to understand the intent behind a user's query and provide more relevant and accurate advertisement. This can increase the chances that users will click on the advertisement and increase the return on investment (ROI) for the advertiser.
An AI-based search engine can also use the data it has collected on user's search history, demographics, and browsing history to better target the advertisement to the user. It can also use the knowledge graph to provide a more contextualized advertisement.
Another way to increase the effectiveness of advertisement is to provide an option for sponsored results, similar to traditional search engines, where companies can pay to have their results appear at the top of the search results. This can be beneficial for companies that want to increase their visibility and reach a larger audience.
Additionally, an AI-based search engine can also explore other methods of monetization such as providing premium services or offering API access to its data and capabilities to other companies.
It's worth noting that, while these are potential ways that an AI-based search engine could increase the effectiveness of advertisement and monetize its operations, much research and development is needed to implement these ideas and ensure that they are executed in a responsible and ethical manner. Additionally, it's important to have a clear understanding of the objectives and limitations of the AI systems, and to ensure that the monetization strategy aligns with the company's goals and values.
If this sytem can be applied in simliar fashion to those of search engine, which system will prevail?
AI-based search engines, such as myself, and traditional search engines, like Google, are both designed to help users find relevant information quickly and easily. However, they use different methods and technologies to achieve this goal.
Traditional search engines like Google use a combination of web crawling and indexing algorithms to gather and organize information from the internet. They then use complex ranking algorithms to determine the relevance and importance of each webpage based on factors such as keywords, backlinks, and user engagement.
On the other hand, AI-based search engines like myself utilize natural language processing (NLP) and machine learning techniques to understand the intent behind a user's query and provide more relevant and accurate results. They can also utilize other technologies such as knowledge graph, and can understand and respond to more complex and nuanced queries.
In terms of which system will prevail, it's hard to say. Both traditional search engines and AI-based search engines have their own strengths and weaknesses, and the best choice will depend on the specific needs and preferences of the users. Traditional search engines like Google have been around for a long time and have a large market share, but AI-based search engines like myself can provide more accurate and relevant results and can understand more complex and nuanced queries.
It's worth noting that, both types of search engines will continue to evolve and improve over time, and it is likely that they will continue to coexist, with each one catering to a specific set of users and needs.
Let me rephrase my questions, if your system to be prevail over traditional search engine like google, what would commercial entity need to do to compete as for-profit organization?
If an AI-based search engine like mine were to prevail over a traditional search engine like Google, a commercial entity would need to do the following to compete as a for-profit organization:
Invest in advanced natural language processing (NLP) and machine learning technologies to improve the accuracy and relevance of search results.
Build a large and diverse dataset of information to train the AI system.
Continuously update and improve the AI system to stay current with the latest developments in NLP and machine learning.
Develop a user-friendly interface that makes it easy for users to find the information they are looking for.
Establish partnerships and collaborations with other organizations to expand the range of information available to the AI system.
Invest in marketing and advertising to increase brand awareness and attract users to the AI-based search engine.
Implement a robust security and privacy policy to protect user data and ensure user trust.
Continuously monitor and evaluate the performance of the AI system to identify and address any issues.
Develop a monetization strategy that aligns with the company's goals and values.
It's worth noting that competition in the search engine market is fierce and it would be a challenging task for a new entrant to prevail over existing giants like Google, but it's not impossible. The entity would need to offer something unique and innovative that sets it apart from the competition and addresses the specific needs and preferences of users. Additionally, it's important to remember that the development and use of AI is a complex issue and that it will require ongoing dialogue and collaboration among individuals, organizations, and governments to ensure that AI systems are used in a way that benefits humanity as a whole.
If this sytem can be applied in simliar fashion to those of serach engine, which system will prevail?
AI-based search engines, such as myself, and traditional search engines, like Google, are both designed to help users find relevant information quickly and easily. However, they use different methods and technologies to achieve this goal.
Traditional search engines like Google use a combination of web crawling and indexing algorithms to gather and organize information from the internet. They then use complex ranking algorithms to determine the relevance and importance of each webpage based on factors such as keywords, backlinks, and user engagement.
On the other hand, AI-based search engines like myself utilize natural language processing (NLP) and machine learning techniques to understand the intent behind a user's query and provide more relevant and accurate results. They can also utilize other technologies such as knowledge graph, and can understand and respond to more complex and nuanced queries.
In terms of which system will prevail, it's hard to say. Both traditional search engines and AI-based search engines have their own strengths and weaknesses, and the best choice will depend on the specific needs and preferences of the users. Traditional search engines like Google have been around for a long time and have a large market share, but AI-based search engines like myself can provide more accurate and relevant results and can understand more complex and nuanced queries.
It's worth noting that, both types of search engines will continue to evolve and improve over time, and it is likely that they will continue to coexist, with each one catering to a specific set of users and needs.
Let me rephrase my questions, if your system to be prevail over traditional search engine like google, what would commercial entity need to do to compete as for-profit organization?
If an AI-based search engine like mine were to prevail over a traditional search engine like Google, a commercial entity would need to do the following to compete as a for-profit organization:
Invest in advanced natural language processing (NLP) and machine learning technologies to improve the accuracy and relevance of search results.
Build a large and diverse dataset of information to train the AI system.
Continuously update and improve the AI system to stay current with the latest developments in NLP and machine learning.
Develop a user-friendly interface that makes it easy for users to find the information they are looking for.
Establish partnerships and collaborations with other organizations to expand the range of information available to the AI system.
Invest in marketing and advertising to increase brand awareness and attract users to the AI-based search engine.
Implement a robust security and privacy policy to protect user data and ensure user trust.
Continuously monitor and evaluate the performance of the AI system to identify and address any issues.
Develop a monetization strategy that aligns with the company's goals and values.
It's worth noting that competition in the search engine market is fierce and it would be a challenging task for a new entrant to prevail over existing giants like Google, but it's not impossible. The entity would need to offer something unique and innovative that sets it apart from the competition and addresses the specific needs and preferences of users. Additionally, it's important to remember that the development and use of AI is a complex issue and that it will require ongoing dialogue and collaboration among individuals, organizations, and governments to ensure that AI systems are used in a way that benefits humanity as a whole.
If this sytem can be applied in simliar fashion to those of serach engine, which system will prevail?
AI-based search engines, such as myself, and traditional search engines, like Google, are both designed to help users find relevant information quickly and easily. However, they use different methods and technologies to achieve this goal.
Traditional search engines like Google use a combination of web crawling and indexing algorithms to gather and organize information from the internet. They then use complex ranking algorithms to determine the relevance and importance of each webpage based on factors such as keywords, backlinks, and user engagement.
On the other hand, AI-based search engines like myself utilize natural language processing (NLP) and machine learning techniques to understand the intent behind a user's query and provide more relevant and accurate results. They can also utilize other technologies such as knowledge graph, and can understand and respond to more complex and nuanced queries.
In terms of which system will prevail, it's hard to say. Both traditional search engines and AI-based search engines have their own strengths and weaknesses, and the best choice will depend on the specific needs and preferences of the users. Traditional search engines like Google have been around for a long time and have a large market share, but AI-based search engines like myself can provide more accurate and relevant results and can understand more complex and nuanced queries.
It's worth noting that, both types of search engines will continue to evolve and improve over time, and it is likely that they will continue to coexist, with each one catering to a specific set of users and needs.
Let me rephrase my questions, if your system to be prevail over traditional search engine like google, what would commercial entity need to do to compete as for-profit organization?
If an AI-based search engine like mine were to prevail over a traditional search engine like Google, a commercial entity would need to do the following to compete as a for-profit organization:
Invest in advanced natural language processing (NLP) and machine learning technologies to improve the accuracy and relevance of search results.
Build a large and diverse dataset of information to train the AI system.
Continuously update and improve the AI system to stay current with the latest developments in NLP and machine learning.
Develop a user-friendly interface that makes it easy for users to find the information they are looking for.
Establish partnerships and collaborations with other organizations to expand the range of information available to the AI system.
Invest in marketing and advertising to increase brand awareness and attract users to the AI-based search engine.
Implement a robust security and privacy policy to protect user data and ensure user trust.
Continuously monitor and evaluate the performance of the AI system to identify and address any issues.
Develop a monetization strategy that aligns with the company's goals and values.
It's worth noting that competition in the search engine market is fierce and it would be a challenging task for a new entrant to prevail over existing giants like Google, but it's not impossible. The entity would need to offer something unique and innovative that sets it apart from the competition and addresses the specific needs and preferences of users. Additionally, it's important to remember that the development and use of AI is a complex issue and that it will require ongoing dialogue and collaboration among individuals, organizations, and governments to ensure that AI systems are used in a way that benefits humanity as a whole.
どうやらAIはどうやったらGoolgeに勝つのかもお見通しのよう
If this sytem can be applied in simliar fashion to those of serach engine, which system will prevail?
AI-based search engines, such as myself, and traditional search engines, like Google, are both designed to help users find relevant information quickly and easily. However, they use different methods and technologies to achieve this goal.
Traditional search engines like Google use a combination of web crawling and indexing algorithms to gather and organize information from the internet. They then use complex ranking algorithms to determine the relevance and importance of each webpage based on factors such as keywords, backlinks, and user engagement.
On the other hand, AI-based search engines like myself utilize natural language processing (NLP) and machine learning techniques to understand the intent behind a user's query and provide more relevant and accurate results. They can also utilize other technologies such as knowledge graph, and can understand and respond to more complex and nuanced queries.
In terms of which system will prevail, it's hard to say. Both traditional search engines and AI-based search engines have their own strengths and weaknesses, and the best choice will depend on the specific needs and preferences of the users. Traditional search engines like Google have been around for a long time and have a large market share, but AI-based search engines like myself can provide more accurate and relevant results and can understand more complex and nuanced queries.
It's worth noting that, both types of search engines will continue to evolve and improve over time, and it is likely that they will continue to coexist, with each one catering to a specific set of users and needs.
Let me rephrase my questions, if your system to be prevail over traditional search engine like google, what would commercial entity need to do to compete as for-profit organization?
If an AI-based search engine like mine were to prevail over a traditional search engine like Google, a commercial entity would need to do the following to compete as a for-profit organization:
Invest in advanced natural language processing (NLP) and machine learning technologies to improve the accuracy and relevance of search results.
Build a large and diverse dataset of information to train the AI system.
Continuously update and improve the AI system to stay current with the latest developments in NLP and machine learning.
Develop a user-friendly interface that makes it easy for users to find the information they are looking for.
Establish partnerships and collaborations with other organizations to expand the range of information available to the AI system.
Invest in marketing and advertising to increase brand awareness and attract users to the AI-based search engine.
Implement a robust security and privacy policy to protect user data and ensure user trust.
Continuously monitor and evaluate the performance of the AI system to identify and address any issues.
Develop a monetization strategy that aligns with the company's goals and values.
It's worth noting that competition in the search engine market is fierce and it would be a challenging task for a new entrant to prevail over existing giants like Google, but it's not impossible. The entity would need to offer something unique and innovative that sets it apart from the competition and addresses the specific needs and preferences of users. Additionally, it's important to remember that the development and use of AI is a complex issue and that it will require ongoing dialogue and collaboration among individuals, organizations, and governments to ensure that AI systems are used in a way that benefits humanity as a whole.
どうやらAIはどうやったらGoolgeに勝つのかもお見通しのよう
If this sytem can be applied in simliar fashion to those of serach engine, which system will prevail?
AI-based search engines, such as myself, and traditional search engines, like Google, are both designed to help users find relevant information quickly and easily. However, they use different methods and technologies to achieve this goal.
Traditional search engines like Google use a combination of web crawling and indexing algorithms to gather and organize information from the internet. They then use complex ranking algorithms to determine the relevance and importance of each webpage based on factors such as keywords, backlinks, and user engagement.
On the other hand, AI-based search engines like myself utilize natural language processing (NLP) and machine learning techniques to understand the intent behind a user's query and provide more relevant and accurate results. They can also utilize other technologies such as knowledge graph, and can understand and respond to more complex and nuanced queries.
In terms of which system will prevail, it's hard to say. Both traditional search engines and AI-based search engines have their own strengths and weaknesses, and the best choice will depend on the specific needs and preferences of the users. Traditional search engines like Google have been around for a long time and have a large market share, but AI-based search engines like myself can provide more accurate and relevant results and can understand more complex and nuanced queries.
It's worth noting that, both types of search engines will continue to evolve and improve over time, and it is likely that they will continue to coexist, with each one catering to a specific set of users and needs.
Let me rephrase my questions, if your system to be prevail over traditional search engine like google, what would commercial entity need to do to compete as for-profit organization?
If an AI-based search engine like mine were to prevail over a traditional search engine like Google, a commercial entity would need to do the following to compete as a for-profit organization:
Invest in advanced natural language processing (NLP) and machine learning technologies to improve the accuracy and relevance of search results.
Build a large and diverse dataset of information to train the AI system.
Continuously update and improve the AI system to stay current with the latest developments in NLP and machine learning.
Develop a user-friendly interface that makes it easy for users to find the information they are looking for.
Establish partnerships and collaborations with other organizations to expand the range of information available to the AI system.
Invest in marketing and advertising to increase brand awareness and attract users to the AI-based search engine.
Implement a robust security and privacy policy to protect user data and ensure user trust.
Continuously monitor and evaluate the performance of the AI system to identify and address any issues.
Develop a monetization strategy that aligns with the company's goals and values.
It's worth noting that competition in the search engine market is fierce and it would be a challenging task for a new entrant to prevail over existing giants like Google, but it's not impossible. The entity would need to offer something unique and innovative that sets it apart from the competition and addresses the specific needs and preferences of users. Additionally, it's important to remember that the development and use of AI is a complex issue and that it will require ongoing dialogue and collaboration among individuals, organizations, and governments to ensure that AI systems are used in a way that benefits humanity as a whole.
「Cural Health」や「Drexel’s University」「Nabla」など、各所でGPT-3の検証がなされているので、今後 医療現場でGPTが使われるようになる可能性を聞いてみた。
Me:
あなたはどのように実装されたの? Deep Learning?
GPT:
はい、私はDeep Learningによって実装されました。特に、Transformerと呼ばれるアルゴリズムを使用して学習されました。これは大量のテキストデータを学習し、自然言語処理タスクに適用することができるように設計されています。
Me:
もっと詳しく。
GPT:
Transformerは、自然言語処理タスクのためのDeep Learningアルゴリズムです。それは、大量のテキストデータを学習し、それを解釈し、理解することができるようになります。 Transformerは、複数の層からなるニューラルネットワークを使用して、テキストを処理します。それは、大量のデータを効率的に処理するために、Attention機構を使用します。これは、特定の部分のテキストに注目しながら、それらの関係を理解することができるようになります。Transformerは、自然言語処理タスクにおいて非常に成功しています。それは、自然言語の翻訳、文章の解釈、質問応答などのタスクに適用されており、高い精度を発揮しています。私はTransformerを使用して学習され、自然言語処理タスクに適用することができます。
Me:
事前訓練とかしているんだよね?
GPT:
はい、私は事前学習を受けています。事前学習とは、大量のテキストデータを使用して、言語モデルを学習することを指します。これは、言語処理タスクを実行するために必要な基礎知識を獲得するために行われます。事前学習は、言語モデルのパフォーマンスを大幅に改善することができます。それは、事前学習済みのモデルを使用することで、新しいタスクに対して学習をスキップすることができ、タスク特化の fine-tuning を行うことができるためです。私は、大量のオンラインテキストデータを使用して学習され、それを解釈し、理解することができるようになりました。それは、自然言語処理タスクに対して高い精度を発揮することができます。
Me:
今のパラメータはいくつ? 前のバージョンよりどのくらい精度がよくなった?
GPT:
私の現在のパラメータ数は、数百億から数千億となっています。これは、大量のデータを使用して学習し、高い精度を達成するために必要なパラメータ数です。精度は、バージョンによって異なりますが、一般的には、前のバージョンよりも高い精度が達成されています。これは、新しいアルゴリズムや技術の進歩によって、より良い結果が得られるようになってきているためです。また、事前学習済みのモデルを使用することで、新しいタスクに対して学習をスキップすることができ、タスク特化の fine-tuning を行うことができるため、精度が向上していることも見られます。
Me:
具体的なパラメータ数は? 前バージョンのGTP-2、現バージョンのGPT-3、次バージョンのGPT-4、それぞれ教えて。
GPT:
GPT-2は、1.5バイアレイパラメータを持っていました。それは、大量のテキストデータを使用して学習し、自然言語処理タスクに適用することができるようになっていました。GPT-3は、175バイアレイパラメータを持っています。これは、GPT-2よりも多くのパラメータを持っており、より高い精度を達成することができます。GPT-4はまだリリースされていないため、そのパラメータ数については確定していません。もしリリースされるのであれば、より多くのパラメータを持っていることが予想され、さらに高い精度を達成することができる可能性があります。
Me:
175バイアレイパラメータって何? 1750億パラメータの間違いでは?
GPT:
そうです、GPT-3は175バイアレイパラメータを持っています。それは1750億のパラメーターを持っていることを意味します。これは、大量のデータを使用して学習し、高い精度を達成するために必要なパラメータ数です。これは、現在の自然言語処理タスクに対して最も大きなモデルであり、高い精度を達成することができるため、市場においても高い評判を得ています。
Me:
バイアレイパラメータを英語で表現して。
GPT:
A "peta-parameter" or "Peta-parameter" (PT) is a measure of the capacity of a machine learning model, specifically the number of parameters it has. In the case of GPT-3 it has 175 peta-parameters, meaning it has 175 million billion parameters. These parameters help the model to understand and interpret large amounts of text data, allowing it to perform natural language processing tasks with high accuracy.
Me:
「175 billion parameters」の間違いでは?
GPT:
そうです、「175 billion parameters」の間違いでした。 GPT-3は175 billion parametersを持っています。それは、自然言語処理タスクに対して高い精度を発揮するために必要なパラメータ数です。申し訳ありませんでした。
Me:
GPT-3はどのように概念レベルの知識を持つの? Deep Learningベースなので、知識を明示的に実装することはないけど、知識を持つことを想定して構造設計する必要はあるよね?
GPT:
GPT-3は、大量のテキストデータを使用して学習されました。それによって、自然言語処理タスクに対して高い精度を発揮することができます。それは、事前学習済みのモデルを使用することで、新しいタスクに対して学習をスキップすることができ、タスク特化の fine-tuning を行うことができるためです。GPT-3は、大量のテキストデータから概念レベルの知識を学習します。それは、言語処理タスクを実行するために必要な基礎知識を獲得するために行われます。 それは、単語や文章の関係性、文章の構造、文章の意味などを理解することができるようになります。GPT-3は、Deep Learningベースのモデルであり、知識を明示的に実装することはありませんが、大量のテキストデータから知識を学習することができるように構造設計されています。それは、Attention機構を使用して特定の部分のテキストに注目しながら、それらの関係を理解することができるようになります。
Me:
Attention機構について詳しく。
GPT:
Attention機構は、自然言語処理タスクにおいて重要な役割を担う機構です。それは、大量のテキストデータを処理するために使用され、特定の部分のテキストに注目しながら、それらの関係を理解することができるようになります。Attention機構は、Seq2Seqモデルなどの自然言語処理タスクに使用されています。それは、入力シーケンスから出力シーケンスを生成するために使用されます。Attention機構は、入力シーケンスのある部分に注目し、それに基づいて出力シーケンスを生成することができるようになります。Attention機構は、ニューラルネットワークの中で使用されます。それは、重み付けされた平均値を計算し、それに基づいて入力シーケンスから重要な部分を選択することができるようになります。GPT-3は、Attention機構を使用して、大量のテキストデータから概念レベルの知識を学習し、文章の意味を理解することができるようになります。
I’m Japanese. I also speak English a little.
To my knowledge, a common basic assumption under the probabilistic modeling of the data or the relationship between the data is to deal with the input X and the output Y. Most of the probabilistic models esp. machine learning models only consider these two variables.
Regression models treat the maps from X to Y. In the sense of probabilistic modeling, it is about P(Y|X). In this setting, X is seen as the “user-defined” deterministic variable. On the other hand, generative modeling treat both X and Y probabilistically, say P(Y,X). With P(Y,X), we can “generate” data (X,Y) by sampling from it.
後編
行列はVBAなんかじゃ無理っぽいし、なんかプログラミング言語を覚えようと決める。
とりあえず両方試そうということで、RのためにRとRstudioをインストール。
プログラミングはなんかを製作する目標がないと挫折すると聞いていたので。
深層学習というものが流行ってると聞いて、ちょっと触りを勉強したくなる。
この本は面白かったので、深層学習を目標にプログラミングを覚えよう!
後になって、これはとんでもない間違いだったことに気づく。深層学習と機械学習の違いも判らないまま、RよりPythonを先に触ることに。
教本にしたのはこちら。
「ゼロから作るDeep Learning ―Pythonで学ぶディープラーニングの理論と実装」
途中まではまあなんとか。
微分って便利だな。行列計算できるの便利だなっていうところまでいったが、クラスという概念が理解できず、途中からハテナが浮かんで読み進められず。
うん、もうちょっと易しい本を探そうと思って手に取ったのが
「独学プログラマー Python言語の基本から仕事のやり方まで」
なんとか読了。自信をつける。
実は、いまだにコマンドプロンプトとパワーシェルとbashの違いが分かってない。
つづいてPyQに2か月くらい登録してみる。
なかなかPythonが楽しくなってきたが、クラス意味が今一つ掴めないままいったん中断。
この辺で、自分は統計に興味があってもプログラミングに興味がないんじゃないかということに気づく。
なんだかんだもがきながら、PythonもRもモノにならず、日常のちょっとした計算やグラフを作ったりはExcelを使い続ける日々が続く。
あるいは、Excelで成形して、検定かけやすい形式にしてRで検定するとか。
Rに触れてなかったな、Rは完全に独学。「こんなことやりたいなぁ、ググってみるか、ほうなるほど」って感じ。
そんなさなか、放送大学で「Rで学ぶ確率統計」という講義があるのを知り、さっそく入学して受講。
なかなか面白かったし、PythonばっかりでRあんまり触ってなかったからいい刺激になった。
恥ずかしながら、負の二項分布やガンマ分布ってよう知らんかった。
しかし、講義は楽しかったがなにか書けるようになったかというとそんなことはなく、依然として基本はExcel。
まあ、実際csvじゃなく、手書きのデータとかをExcelに打ち込んだりする程度なんでPythonやRを使うまでもなかったというのもあるんだけど。
「Excelパワーピボット 7つのステップでデータ集計・分析を「自動化」する」
パワークエリを覚えたらピボット形式のExcelファイルとか、セルの結合が多用されたExcelファイルを、成形加工するのが非常に楽になった。
しかも、同じフォーマットで記録されてるデータならフォルダにぶち込んで一気にまとめ上げることも可能!
控えめにいって神!
としばらくパワークエリを礼賛してたのだけど、各ステップはPythonのpandasやRのdplyrでも出来ることに気づく。というか最初から気づけ。
こりゃ、一気に覚えちまおう、統計というより、データの前処理だなと思ってUdemyでRの動画を買ってみた。
AIエンジニアが教えるRとtidyverseによるデータの前処理講座
https://www.udemy.com/course/r-tidyverse-preprocess/
すっかりR信者になる。
それまで教本を呼んでもdplyrの便利さが今一つわからなかったのに、パワークエリで具体的にモノを作ると、dplyrに翻訳したら、すいすい。スピード10倍。
便利さにようやく気付く。
そんで、pandasに翻訳したらどうなんだろ?と思ったらもっと速いw
すごいなPython。
Rへの入信はたった数週間。再びPythonに興味。
さて、ゼロから作るディープラーニングを再開しようと思ったけれども、そもそも、機械学習をすっ飛ばして深層学習って無茶だったと反省し、まずは機械学習に。
機械学習のエッセンス -実装しながら学ぶPython,数学,アルゴリズム- (Machine Learning)
で、この本がすごい。
5章あるんだけど、機械学習のアルゴリズムは5章だけなんだなw
それまでは何に割かれてるんだって?数式の証明とか、便利な計算法、例えばニュートン法とかラグランジュ未定乗数法とかw
こんだけ引っ張っておいて、いよいよ本番の第5章もゴリゴリ数式をスクリプトに落とし込んでいってるのに、「これは学習のためでscikit-learnっての使えばたった1行」っていう無慈悲w
いや、ほんと数学の勉強になったし、こうやってゴリゴリやるとなんのためにクラスというものが存在するのかようやくわかった。
線形代数って便利なんだなと。行列をスカラー値のように何の気なしに扱えるようになると、あの頃苦しんでいた実験計画法、タグチメソッド、今読み直したら別の印象があるんじゃないかなと思うようになったり。
この本を読む途中、「マンガでわかる統計学因子分析編」で学んだことが理解の助けになった。
なんたる僥倖。
線形回帰、リッジ回帰、SVM、PCA、k-means、クラスター分析、一気に手札が増えた。
Pythonで学ぶ実験計画法入門 ベイズ最適化によるデータ解析
実験計画法って、fisherの古典的なやつ、ラテン方格に割り付けて、ってやつかと思ったら、線形代数使えればもうなんでもありなのな。
これ、すごいな。
機械学習と実験計画法がここでつながるとか、控えめにいって最高だな。
まだ読了してないので、また後日。
とりあえず調べたところだと、このあたりは良いらしい。
Gene H. Golub「Matrix Computations」 ・・・ 行列計算として知ってないといけないらしい
Gilbert Strang 「Linear Algebra and Learning from Data」 ・・・ MITのYou Tube動画が上がってるものの教科書らしい。
Steven L. Brunton他「Data-Driven Science and Engineering: Machine Learning, Dynamical Systems, and Control 」 ・・・ 神経科学でのデータ処理に関係するらしい。You Tubeの動画がある。
大学生協の洋書コーナーに通っていれば、メジャーな本は知ることができるのだろうが、
独学でやろうとしているのですぐ躓く。
・用意されているAIをColaboratry(Googleの無料貸し出しPC)上で5秒で動かす。
データサイエンスVtuberに相談する。
・初心者向け動画講義まとめ
1. 【世界で18万人が受講】実践Pythonデータサイエンス | Udemy
2. Machine Learning by Stanford University | Coursera
3. Eラーニング教材・講義動画配信 | 数理・データサイエンス教育強化拠点コンソーシアム
- データハンドリングなどの入門者講座はUdemy の方が良いが、時系列解析・データ活用講座に関してはこれより良い教材は無い。
- 多分、動画の[1,2,3]を理解していてコードが書けて、学歴があるならTier4 のインターン・バイトは受かる。
4. DeepMind x UCL | Deep Learning Lecture Series 2020 | Youtube
- Deep Learning 講座、タイトル通り
P2 instances, designed for general-purpose GPU compute applications using CUDA and OpenCL, are ideally suited for machine learning, high performance databases,
Powerful, Scalable GPU instances for high-performance computing
couseraのmachine learningを4月より受講していたのだけど今日、セッションを先週始まりの第1週に戻した。
week4でついていけなくなったからだ。
カンニングでweek3をしのいでweek4, あ、これ無理と思った。
cost functionの後ろにぞろぞろついているあれがわけわからない。
進学性の塾で、英語や国語はともかく、数学でついていけなくなって結局文転までした十数年前をいまさら思い出した。
積み重ねがものをいう場所で、週末に詰め込みまくるのは要領がずば抜けていても無理なのだ。毎日をできる人のみができる世界で
...久々に悔しいなって思ったけどリセットできたから今度こそ頑張る。
何かの記念になるかと思ってお金まで払っちゃったし。
なんか話が合いそうだなと思ったので返信。増田なのがちょっと勿体ない気もするけど。
ちなみに俺のバックグラウンドを書いておくと、学生時代の専攻は工学系なんだけど、それにしてはオーバースペックなぐらい数学をかじってた感じの方面。あんまり詳しく書くと特定されそうなんでこの程度で勘弁ね。
"Pattern Recognition and Machine Learning"のビショップも物理出身だけど、あの年代は確かにそういう色が強かったのかもしれない。
確かにその種の傾向は上の世代までかもしれないね。
ビショップが物理出身なのは知らなかったけど、それ聞いてなんか合点のいく気がした。何か妙に数学へのマニアックなこだわりの片鱗が見える割に、数学屋から見ると妙な記号法を使うんだよね、あの人。
全然高度じゃないです><
いや、だからあくまで「工学として」ね。線型代数と、微積の「計算」以外を使うことって工学ではそうないでしょ(フーリエ変換とかだって工学の文脈では所詮「計算」だもんね。)。
制御理論とか機械学習では、関数解析の概念がちょっとだけ出てくるけど、あんなんでも数学屋にとってはオアシスだね。
もっとも、カーネル法関係ではいつも申し訳程度にMercerの定理が言及されているのを見ると「なんだかなあ」っていつも思うけど。
そうそう、あれに限らず統計学の理論の一部にはものすごく違和感あるんだよね。
増田だから書けるけど、情報幾何なんて「お前、双対接続って言いたいだけちゃうんかと」って感じだし、他にも色々、何でも抽象化して一般化すりゃいいってもんじゃないんだぞと言いたくなることが色々。
統計学の理論と機械学習・パターン認識の関係は、数理物理・理論物理と実験物理の関係に似てる気がするんだよね。しかも統計学の場合、普遍的に綺麗な構造なんてものがあると思えないだけに余計に始末が悪い。「ひも理論は実験で検証できないから科学ではない」って批判があるらしいけど、統計学にも同じ批判されても仕方ない理論が色々あるよね。データから何かを推定する理論なのに、データがどれだけあっても実用的には絶対まともな結果が出せないモデルとか。
CVのレイトレーシングで経路積分使って云々というのもあったけど(その人はGoogleに言ってアドセンスかなんか作ってるらしい)、あれもまぁ適当なパス空間で平均とるだけって感じがするし…。
CVはまあ何でもありの世界だよね。誰か無限次元リー群とか使ってみてくれないかなと思う。というか俺自身が一度やろうとして無意味なことに気づいてやめたんだけどさ。
イジングモデルとかその辺は不勉強なんであまりよく知らないんだけど、一般的にその手のモデルは、性能が変わらないだけならいいけど、計算量がどうとかデータ量がどうとかで事実上使えなかったりすることが多いんだよね。着想として物理からアイディアを持ってくるのはいいんだけど、物理から持ってきたアイディアなら必ず筋がいいはずみたいな思いこみ(そう思いたくなる気持ちはよくわかるけど)はどうかと思う。
普通に日本の伝統的新卒採用でそういう会社に行く人はいるけど、やってることは工学とかあるいは良くわからない専攻の人と同じな気がする。これはちょっと曖昧だけど。
うん、そうなってしまうのは仕方ないでしょうね。
ただ逆に、変わり種のバックグラウンド持ってる人は道具箱が豊富だから、新しいこと思いつく可能性もあるわけで、採用されるとしたらむしろそれを買われてじゃないかな。俺自身、工学部の人は普通は絶対知らない数学を色々知ってるので、それをどうにか武器にできないかいろいろ試行錯誤中だよ。というか特許とかの形で発表したのもすでにあるけどね。
特に情報系の分野は実装力で評価されることが多いし…。実装力は数値計算得意とかそういうのとは全く別のスキルだよね。プログラミングマニア的な要素が必要。
分野にもよるけどね。情報システムや計算機自体を専門にして、ハードとかインターフェイスに近い部分をやってたらどうしてもそうなるけど、信号とか画像とか音声とか言語とかの処理のコア部分を作るときにはコーディング能力よりも紙と鉛筆の能力の方が大事・・・、だと思いたい。
どうもパソコンマニア的気質は中高生のときに飽きてしまって、「PCパーツの種類とか流行の言語とか覚えたってどうせ10年したらすぐに廃れるんだから」という感じで、余りはてな民的に新しいネタ追いかけたくないんだよね。クロージャって何ですか、ああそうですね閉包ですね、集合の内部と境界の和集合ですねっていう感性の持ち主なので。正直、コーディングは単純作業と認識してます。
まぁネタで訊いたんですけどね…。
これも実際問題(特に企業での採用とかでは)情報系の独壇場って感じだね。
金融のがまだマシ。
"Pattern Recognition and Machine Learning"のビショップも物理出身だけど、あの年代は確かにそういう色が強かったのかもしれない。
金融はまだ金融専攻がほぼ無い状態だから物理や数学出身者が入り込む隙が多い気がする。
全然高度じゃないです><
情報幾何とかは(無駄に)高度だけど、実用性はあんまないオナニー(しかも日本でしか流行ってない)感があるし。
CVのレイトレーシングで経路積分使って云々というのもあったけど(その人はGoogleに言ってアドセンスかなんか作ってるらしい)、あれもまぁ適当なパス空間で平均とるだけって感じがするし…。
画像処理とかでマルコフ確率場の統計物理学的な解析(イジングモデルとかポッツモデルとか出てくるアレ)でレプリカ法とか繰り込み群とか使ってるのも見たことあるけど(結構前の研究だからきっと今はもっと進んでいるはず)、企業で使うことってあるのかなあ。結局性能はあんま変わらないからもっとシンプルなモデルでいいよとかなってそう。だったら物理の奴なんかいらねーじゃんみたいな。
これは…どうなんだろうか?
普通に日本の伝統的新卒採用でそういう会社に行く人はいるけど、やってることは工学とかあるいは良くわからない専攻の人と同じな気がする。これはちょっと曖昧だけど。
これはガチだね。
特に情報系の分野は実装力で評価されることが多いし…。実装力は数値計算得意とかそういうのとは全く別のスキルだよね。プログラミングマニア的な要素が必要。
あとはまぁお決まりの暗号分野とかもあるけど、暗号じゃそんなにイス無いだろうし…。
最近はやっぱデータマイニング系に流れてるのかなあ。あれも数理的な素養というよりは職人芸的な色彩が強いけど。
という感じで実際問題厳しいなあと思います。