MY-ASSIGNMENTEXPERT™可以为您提供sydney COMP5318 Machine Learning机器学习课程的代写代考和辅导服务！
这是悉尼大学机器学习课程的代写成功案例。
COMP5318课程简介
Machine learning is the process of automatically building mathematical models that explain and generalise datasets. It integrates elements of statistics and algorithm development into the same discipline. Data mining is a discipline within knowledge discovery that seeks to facilitate the exploration and analysis of large quantities for data, by automatic and semiautomatic means. This subject provides a practical and technical introduction to machine learning and data mining. Topics to be covered include problems of discovering patterns in the data, classification, regression, feature extraction and data visualisation. Also covered are analysis, comparison and usage of various types of machine learning techniques and statistical techniques.
Prerequisites
At the completion of this unit, you should be able to:
- LO1. understand the basic principles, strengths, weaknesses and applicability of machine learning algorithms for solving classification, regression, clustering and reinforcement learning tasks.
- LO2. have obtained practical experience in designing, implementing and evaluating machine learning algorithms
- LO3. have gained practical experience in using machine learning software and libraries
- LO4. present and interpret data and information in verbal and written form
COMP5318 Machine Learning HELP（EXAM HELP， ONLINE TUTOR）
Which of the following describes an machine learning approach to build a system for spam detection? Choose the correct answer; explain briefly why you think other choices are not machine learning.
[a] flip 3 fair coins; classify the email as a spam iff at least 2 of them are heads
[b] forward the email to 3 humans; classify the email as a spam iff at least 2 of them believe so
[c] produce a list of words for spams by 3 humans; classify the email as a spam iff the email contains more than 10 words from the list
[d] get a data set that contains spams and non-spams, for all words in the data set, let the machine calculate the ratio of spams per word; produce a list of words that appear more than 5 times and are of the highest $20 \%$ ratio; classify the email as a spam iff the email contains more than 10 words from the list
[e] get a data set that contains spams and non-spams, for all words in the data set, let the machine decide its “spam score”; sum the score up for each email; let the machine optimize a threshold that achieves the best precision of spam detection; classify the email as a spam iff the email is of score more than the threshold
The scaling in the previous problem is equivalent to inserting a “learning rate” $\eta$ to the PLA update rule
$$
\mathbf{w}{t+1} \leftarrow \mathbf{w}_t+\eta \cdot y{n(t)} \mathbf{x}{n(t)} $$ with $\eta=\frac{1}{4}$. In fact, we do not need to use a fixed $\eta$. Let $\eta_t$ denote the learning rate in the $t$-th iteration; that is, let PLA update $\mathbf{w}_t$ by $$ \mathbf{w}{t+1} \leftarrow \mathbf{w}t+\eta_t \cdot y{n(t)} \mathbf{x}{n(t)} $$ whenever $\left(\mathbf{x}{n(t)}, y_{n(t)}\right)$ is not correctly classified by $\mathbf{w}t$. Dr. Adaptive decides to set $\eta_t=\frac{1}{\left|\mathbf{x}{n(t)}\right|}$ so “longer” $\mathbf{x}{n(t)}$ will not affect $\mathbf{w}_t$ too much. Let $$ \hat{\rho}=\min {n \in{1,2, \cdots, N}} \frac{\left|\mathbf{w}_f^T \mathbf{x}_n\right|}{\left|\mathbf{w}_f\right|\left|\mathbf{x}_n\right|},
$$
which can be viewed as a “normalized” version of the $\rho$ on page 16 of lecture 2 . The bound on the same page then becomes $\hat{\rho}^{-p}$ after using this adaptive $\eta_t$. What is $p$ ? Choose the correct answer; explain your answer.
[a] 0
[b] 1
[c] 2
[d] 4
[e] 8
Dr. Separate decides to use one of the update rules in the previous problem for PLA. When the data set is linear separable, how many choices in the previous problem ensures halting with a “perfect line”? Choose the correct answer; explain the reason behind each halting case.
[a] 1
[b] 2
[c] 3
[d] 4
[e] 5
One shared technique between the famous AlphaGo, AlphaGo Zero, and AlphaStar is called selfpracticing: learning to play the game by practicing with itself and getting the feedback from the “judge” environment. What best describes the learning problem behind self-practicing? Choose the correct answer; explain your answer.
[a] human learning
[b] unsupervised learning
[c] semi-supervised learning
[d] supervised learning
[e] reinforcement learning
Consider formulating a learning problem for building a self-driving car. First, we gather a training data set that consists of 100 hours of video that contains the view in front of a car, and records about how the human behind the wheel acted with physically constrained choices like steering, braking, and signaling-before-turning. We also gather another 100 hours of videos from 1126 more cars without the human records. The learning algorithm is expected to learn from all the videos to obtain a hypothesis that imitates the human actions well. What learning problem best matches the description above? Choose the correct answer; explain your answer.
[a] regression, unsupervised learning, active learning, concrete features
[b] structured learning, semi-supervised learning, batch learning, raw features
[c] structured learning, supervised learning, batch learning, concrete features
[d] regression, reinforcement learning, batch learning, concrete features
[e] structured learning, supervised learning, online learning, concrete features
(We are definitely not hinting that you should build a self-driving car this way. 😀 )
MY-ASSIGNMENTEXPERT™可以为您提供SYDNEY COMP5318 MACHINE LEARNING机器学习课程的代写代考和辅导服务！
