Personal AI Machine Learning Tutor Online That Teaches the Method
Describe the model and the tutor asks whether the error is coming from the model being too simple or too flexible, because that single distinction explains most of what a machine learning course is teaching.
What is the AskSia AI machine learning tutor?
The AskSia AI machine learning tutor is a personal tutoring tool that teaches why models fail rather than which library function to call. It asks whether error is coming from bias or variance, walks the loss function or the optimisation step with you when you ask, and asks what would change if the dataset were larger. It covers supervised learning, loss functions and optimisation, regularisation, tree and ensemble methods, unsupervised learning, neural network fundamentals, validation strategy and evaluation. It answers in eight languages. Students on the AskSia A+ plan are covered by the Grade Confidence Guarantee. AskSia is a study aid, not an answer key.
如同真人導師的四個階段
解答出現時，解題工具就完成任務；當你能獨自解下一題，輔導才算完成。
給推進方向，而不是直接給解答
它指出卡住的步驟，再把問題交回給你走出下一步。
逐行說明，搭配圖解
需要更多協助時，它會逐行帶你走過解法，而不只告訴你最終結果。
它會反過來問你一題
在繼續之前，它會用一個簡短問題確認下一步是否真正屬於你。
稍後再回來複習
弱項會在更適合主動回憶、鞏固解題方法的時機再次出現。
考試還有三週
導師會刻意放慢節奏，為下一題留下可重複使用的解題方法。
每學期最容易失分的三個習慣
Machine learning coursework is graded on diagnosis. These three are where students stall.
Adding complexity to fix a variance problem
When training error is low and validation error is high, a bigger model makes it worse. The tutor asks which error is high before anything is changed.
Tuning hyperparameters on the test set
Using test performance to choose settings turns it into a second training set and the reported result becomes meaningless. The tutor asks which split each decision used.
Reading a cluster as a category
Clustering finds structure in the chosen features and distance metric, not real world groups. The tutor asks what a different feature set would have produced.
不是題庫，而是從你的錯誤建立的測驗
Your marks are not lost evenly, so your practice is not spread evenly. Try this one.
這位導師涵蓋的內容
A first machine learning course, organised around why models behave the way they do.
Missing the maths underneath? Linear algebra, calculus and probability carry most of the derivations. Working on the applied pipeline instead?
Four things a machine learning tutor does that a chat box does not
它了解你的學習歷程，不只是一個問題
下一次輔導會從你真正感到困難的步驟開始，而不是泛泛地重講整個主題。
它會畫出圖解，不只列出代數式
圖解會帶著標記與解題過程的每一步並列呈現，因為視覺化推理本身也能得到分數。
它會找出你自己解題過程中的第一行錯誤
拍下做錯的嘗試，它會找到解法最早偏離的步驟，接著講清這一步。
它依照你的課程方式檢驗理解
長答題需要依照你自己試卷的格式練習，而不是套用選擇題題庫。
Why diagnosis is the whole subject
Anyone can fit a model in three lines. What a course examines is whether you can look at two error numbers and say what to change. The tutor asks which error is high before it discusses any adjustment, because that ordering is the method.
Personal here means it remembers which diagnosis keeps failing. If your implementations are clean but validation discipline keeps slipping, that is tracked and raised before the next assignment.
It works in English, German, Japanese, Korean, Spanish, Portuguese, Simplified Chinese and Traditional Chinese.
學生真正會問的問題
Will it write the model code?
No. It asks which error is high and what that implies, then walks the reasoning. For implementation help, use the Python page.
Can it help me diagnose overfitting?
Yes, and it is the most requested topic here. It starts from your two error numbers rather than from a definition.
Does it cover neural networks?
Yes, at the level a first course teaches: layers, activations, backpropagation and why depth helps. Framework specific tutorials are outside its scope.
Can it explain a derivation?
Yes, including gradient descent and the regularised objectives, and it will ask which step you are stuck on rather than restating the whole derivation.
Does it cover unsupervised methods?
Yes, including clustering and dimensionality reduction, with emphasis on what those results do and do not mean.
Is using an AI machine learning tutor allowed?
AskSia is built as a study aid, not an answer key. Check your own faculty's policy before using any tool on graded work.
全部 149 位 AskSia 導師
每個學科都採用同一種方式：針對你卡住的步驟給出提示，結合你自己的課程資料講解，最後再向你拋回一個問題。
數學和統計學
15從代數到微積分，也提供面對截止日的作業協助。
科學
11物理、化學與生命科學的導師頁面和作業協助。
計算和數據
22涵蓋主要程式語言，以及課程實際使用的理論、資料與工具。
工程
8從靜力學到流體力學，並提供應用課程作業頁面。
商業和經濟
9會計、金融與經濟，搭配圖表及財務報表。
人文與寫作
13協助檢閱草稿而非代寫的人文與寫作科目。
語言
16提供附規則的修正，以及目標語言的對話練習。
AP課程
20依照內容與評分規準共同設計。
考試和認證
11以實際評分格式練習證照與入學考試。
健康相關課程
7結合每個選項的理據，協助課程與執照準備。
學習支持
9當你需要重整整份行程，或今晚就要面對截止日。
比較學習工具
8比較各工具的強項，以及它們無法處理的範圍。
