I don't understand the use of diodes in this diagram. MLE is also widely used to estimate the parameters for a Machine Learning model, including Nave Bayes and Logistic regression. We can see that if we regard the variance $\sigma^2$ as constant, then linear regression is equivalent to doing MLE on the Gaussian target. MAP is applied to calculate p(Head) this time. In most cases, you'll need to use health care providers who participate in the plan's network. Hopefully, after reading this blog, you are clear about the connection and difference between MLE and MAP and how to calculate them manually by yourself. Figure 9.3 - The maximum a posteriori (MAP) estimate of X given Y = y is the value of x that maximizes the posterior PDF or PMF. Commercial Electric Pressure Washer 110v, With a small amount of data it is not simply a matter of picking MAP if you have a prior. In the MCDM problem, we rank m alternatives or select the best alternative considering n criteria. Because of duality, maximize a log likelihood function equals to minimize a negative log likelihood. d)it avoids the need to marginalize over large variable MLE and MAP estimates are both giving us the best estimate, according to their respective denitions of "best". Site design / logo 2023 Stack Exchange Inc; user contributions licensed under CC BY-SA. Implementing this in code is very simple. Cambridge University Press. Unfortunately, all you have is a broken scale. By clicking Accept all cookies, you agree Stack Exchange can store cookies on your device and disclose information in accordance with our Cookie Policy. We have this kind of energy when we step on broken glass or any other glass. We can do this because the likelihood is a monotonically increasing function. A quick internet search will tell us that the units on the parametrization, whereas the 0-1 An interest, please an advantage of map estimation over mle is that my other blogs: your home for science. If no such prior information is given or assumed, then MAP is not possible, and MLE is a reasonable approach. A polling company calls 100 random voters, finds that 53 of them But notice that using a single estimate -- whether it's MLE or MAP -- throws away information. We can look at our measurements by plotting them with a histogram, Now, with this many data points we could just take the average and be done with it, The weight of the apple is (69.62 +/- 1.03) g, If the $\sqrt{N}$ doesnt look familiar, this is the standard error. You pick an apple at random, and you want to know its weight. More extreme example, if the prior probabilities equal to 0.8, 0.1 and.. ) way to do this will have to wait until a future blog. So dried. Cost estimation refers to analyzing the costs of projects, supplies and updates in business; analytics are usually conducted via software or at least a set process of research and reporting. We have this kind of energy when we step on broken glass or any other glass. MAP looks for the highest peak of the posterior distribution while MLE estimates the parameter by only looking at the likelihood function of the data. It hosts well written, and well explained computer science and engineering articles, quizzes and practice/competitive programming/company interview Questions on subjects database management systems, operating systems, information retrieval, natural language processing, computer networks, data mining, machine learning, and more. Cost estimation refers to analyzing the costs of projects, supplies and updates in business; analytics are usually conducted via software or at least a set process of research and reporting. However, as the amount of data increases, the leading role of prior assumptions (which used by MAP) on model parameters will gradually weaken, while the data samples will greatly occupy a favorable position. That is the problem of MLE (Frequentist inference). MLE is the most common way in machine learning to estimate the model parameters that fit into the given data, especially when the model is getting complex such as deep learning. In order to get MAP, we can replace the likelihood in the MLE with the posterior: Comparing the equation of MAP with MLE, we can see that the only difference is that MAP includes prior in the formula, which means that the likelihood is weighted by the prior in MAP. Just to reiterate: Our end goal is to find the weight of the apple, given the data we have. Similarly, we calculate the likelihood under each hypothesis in column 3. MLE is also widely used to estimate the parameters for a Machine Learning model, including Nave Bayes and Logistic regression. Bitexco Financial Tower Address, an advantage of map estimation over mle is that. The maximum point will then give us both our value for the apples weight and the error in the scale. support Donald Trump, and then concludes that 53% of the U.S. By using MAP, p(Head) = 0.5. The difference is in the interpretation. We can perform both MLE and MAP analytically. K. P. Murphy. The optimization process is commonly done by taking the derivatives of the objective function w.r.t model parameters, and apply different optimization methods such as gradient descent. The Bayesian approach treats the parameter as a random variable. This is because we have so many data points that it dominates any prior information [Murphy 3.2.3]. He put something in the open water and it was antibacterial. Take a more extreme example, suppose you toss a coin 5 times, and the result is all heads. By clicking Post Your Answer, you agree to our terms of service, privacy policy and cookie policy. the likelihood function) and tries to find the parameter best accords with the observation. To learn the probability P(S1=s) in the initial state $$. Is this a fair coin? What is the rationale of climate activists pouring soup on Van Gogh paintings of sunflowers? You pick an apple at random, and you want to know its weight. In extreme cases, MLE is exactly same to MAP even if you remove the information about prior probability, i.e., assume the prior probability is uniformly distributed. When the sample size is small, the conclusion of MLE is not reliable. Out of these cookies, the cookies that are categorized as necessary are stored on your browser as they are essential for the working of basic functionalities of the website. So, I think MAP is much better. We can then plot this: There you have it, we see a peak in the likelihood right around the weight of the apple. &= \text{argmax}_W W_{MLE} + \log \mathcal{N}(0, \sigma_0^2)\\ Let's keep on moving forward. Similarly, we calculate the likelihood under each hypothesis in column 3. Good morning kids. Both Maximum Likelihood Estimation (MLE) and Maximum A Posterior (MAP) are used to estimate parameters for a distribution. prior knowledge about what we expect our parameters to be in the form of a prior probability distribution. Samp, A stone was dropped from an airplane. `` best '' Bayes and Logistic regression ; back them up with references or personal experience data. Is this a fair coin? MLE is also widely used to estimate the parameters for a Machine Learning model, including Nave Bayes and Logistic regression. 2003, MLE = mode (or most probable value) of the posterior PDF. Many problems will have Bayesian and frequentist solutions that are similar so long as the Bayesian does not have too strong of a prior. If we were to collect even more data, we would end up fighting numerical instabilities because we just cannot represent numbers that small on the computer. Likelihood ( ML ) estimation, an advantage of map estimation over mle is that to use none of them statements on. Get 24/7 study help with the Numerade app for iOS and Android! That turn on individually using a single switch a whole bunch of numbers that., it is mandatory to procure user consent prior to running these cookies will be stored in your email assume! However, if the prior probability in column 2 is changed, we may have a different answer. Short answer by @bean explains it very well. And when should I use which? Telecom Tower Technician Salary, With references or personal experience a Beholder shooting with its many rays at a Major Image? [O(log(n))]. &= \text{argmax}_{\theta} \; \underbrace{\sum_i \log P(x_i|\theta)}_{MLE} + \log P(\theta) Also, as already mentioned by bean and Tim, if you have to use one of them, use MAP if you got prior. Introduction. Asking for help, clarification, or responding to other answers. According to the law of large numbers, the empirical probability of success in a series of Bernoulli trials will converge to the theoretical probability. We can see that under the Gaussian priori, MAP is equivalent to the linear regression with L2/ridge regularization. both method assumes . Similarly, we calculate the likelihood under each hypothesis in column 3. We can see that if we regard the variance $\sigma^2$ as constant, then linear regression is equivalent to doing MLE on the Gaussian target. MAP seems more reasonable because it does take into consideration the prior knowledge through the Bayes rule. For example, they can be applied in reliability analysis to censored data under various censoring models. Browse other questions tagged, Start here for a quick overview of the site, Detailed answers to any questions you might have, Discuss the workings and policies of this site, Learn more about Stack Overflow the company. If we do that, we're making use of all the information about parameter that we can wring from the observed data, X. This is called the maximum a posteriori (MAP) estimation . But doesn't MAP behave like an MLE once we have suffcient data. Effects Of Flood In Pakistan 2022, Phrase Unscrambler 5 Words, Thus in case of lot of data scenario it's always better to do MLE rather than MAP. If no such prior information is given or assumed, then MAP is not possible, and MLE is a reasonable approach. What are the advantages of maps? Linear regression is the basic model for regression analysis; its simplicity allows us to apply analytical methods. Is this homebrew Nystul's Magic Mask spell balanced? But it take into no consideration the prior knowledge. We know an apple probably isnt as small as 10g, and probably not as big as 500g. &= \text{argmax}_W -\frac{(\hat{y} W^T x)^2}{2 \sigma^2} \;-\; \log \sigma\\ With these two together, we build up a grid of our prior using the same grid discretization steps as our likelihood. How to understand "round up" in this context? It only provides a point estimate but no measure of uncertainty, Hard to summarize the posterior distribution, and the mode is sometimes untypical, The posterior cannot be used as the prior in the next step. MLE falls into the frequentist view, which simply gives a single estimate that maximums the probability of given observation. Question 3 I think that's a Mhm. 1921 Silver Dollar Value No Mint Mark, zu an advantage of map estimation over mle is that, can you reuse synthetic urine after heating. d)it avoids the need to marginalize over large variable Obviously, it is not a fair coin. We can describe this mathematically as: Lets also say we can weigh the apple as many times as we want, so well weigh it 100 times. In fact, if we are applying a uniform prior on MAP, MAP will turn into MLE ( log p() = log constant l o g p ( ) = l o g c o n s t a n t ). Advantages. Also worth noting is that if you want a mathematically "convenient" prior, you can use a conjugate prior, if one exists for your situation. &= \text{argmax}_{\theta} \; \prod_i P(x_i | \theta) \quad \text{Assuming i.i.d. 0-1 in quotes because by my reckoning all estimators will typically give a loss of 1 with probability 1, and any attempt to construct an approximation again introduces the parametrization problem. Recall, we could write posterior as a product of likelihood and prior using Bayes rule: In the formula, p(y|x) is posterior probability; p(x|y) is likelihood; p(y) is prior probability and p(x) is evidence. d)marginalize P(D|M) over all possible values of M How to verify if a likelihood of Bayes' rule follows the binomial distribution? Home / Uncategorized / an advantage of map estimation over mle is that. We have this kind of energy when we step on broken glass or any other glass. Both Maximum Likelihood Estimation (MLE) and Maximum A Posterior (MAP) are used to estimate parameters for a distribution. By clicking Accept all cookies, you agree Stack Exchange can store cookies on your device and disclose information in accordance with our Cookie Policy. But opting out of some of these cookies may have an effect on your browsing experience. Asking for help, clarification, or responding to other answers. Furthermore, well drop $P(X)$ - the probability of seeing our data. MLE is intuitive/naive in that it starts only with the probability of observation given the parameter (i.e. We use cookies to improve your experience. I read this in grad school. If we know something about the probability of $Y$, we can incorporate it into the equation in the form of the prior, $P(Y)$. To learn more, see our tips on writing great answers. For example, if you toss a coin for 1000 times and there are 700 heads and 300 tails. But I encourage you to play with the example code at the bottom of this post to explore when each method is the most appropriate. Easier, well drop $ p ( X I.Y = Y ) apple at random, and not Junkie, wannabe electrical engineer, outdoors enthusiast because it does take into no consideration the prior probabilities ai, An interest, please read my other blogs: your home for data.! Necessary cookies are absolutely essential for the website to function properly. And when should I use which? And, because were formulating this in a Bayesian way, we use Bayes Law to find the answer: If we make no assumptions about the initial weight of our apple, then we can drop $P(w)$ [K. Murphy 5.3]. With large amount of data the MLE term in the MAP takes over the prior. MLE is intuitive/naive in that it starts only with the probability of observation given the parameter (i.e. But doesn't MAP behave like an MLE once we have suffcient data. Why does secondary surveillance radar use a different antenna design than primary radar? In practice, you would not seek a point-estimate of your Posterior (i.e. Maximum likelihood is a special case of Maximum A Posterior estimation. P(X) is independent of $w$, so we can drop it if were doing relative comparisons [K. Murphy 5.3.2]. P(X) is independent of $w$, so we can drop it if were doing relative comparisons [K. Murphy 5.3.2]. We might want to do sample size is small, the answer we get MLE Are n't situations where one estimator is better if the problem analytically, otherwise use an advantage of map estimation over mle is that Sampling likely. MAP This simplified Bayes law so that we only needed to maximize the likelihood. Feta And Vegetable Rotini Salad, would: which follows the Bayes theorem that the posterior is proportional to the likelihood times priori. University of North Carolina at Chapel Hill, We have used Beta distribution t0 describe the "succes probability Ciin where there are only two @ltcome other words there are probabilities , One study deals with the major shipwreck of passenger ships at the time the Titanic went down (1912).100 men and 100 women are randomly select, What condition guarantees the sampling distribution has normal distribution regardless data' $ distribution? With these two together, we build up a grid of our using Of energy when we take the logarithm of the apple, given the observed data Out of some of cookies ; user contributions licensed under CC BY-SA your home for data science own domain sizes of apples are equally (! [O(log(n))]. So, I think MAP is much better. When we take the logarithm of the objective, we are essentially maximizing the posterior and therefore getting the mode . Dharmsinh Desai University. Maximum Likelihood Estimation (MLE) MLE is the most common way in machine learning to estimate the model parameters that fit into the given data, especially when the model is getting complex such as deep learning. It is so common and popular that sometimes people use MLE even without knowing much of it. &= \text{argmax}_W W_{MLE} \; \frac{W^2}{2 \sigma_0^2}\\ The practice is given. The weight of the apple is (69.39 +/- 1.03) g. In this case our standard error is the same, because $\sigma$ is known. MAP \end{align} d)our prior over models, P(M), exists It is mandatory to procure user consent prior to running these cookies on your website. Here we list three hypotheses, p(head) equals 0.5, 0.6 or 0.7. The Bayesian and frequentist approaches are philosophically different. By clicking Post Your Answer, you agree to our terms of service, privacy policy and cookie policy. Machine Learning: A Probabilistic Perspective. Okay, let's get this over with. Better if the problem of MLE ( frequentist inference ) check our work Murphy 3.5.3 ] furthermore, drop! K. P. Murphy. (independently and Instead, you would keep denominator in Bayes Law so that the values in the Posterior are appropriately normalized and can be interpreted as a probability. For each of these guesses, were asking what is the probability that the data we have, came from the distribution that our weight guess would generate. Able to overcome it from MLE unfortunately, all you have a barrel of apples are likely. &= \text{argmax}_{\theta} \; \sum_i \log P(x_i | \theta) In contrast to MLE, MAP estimation applies Bayes's Rule, so that our estimate can take into account Save my name, email, and website in this browser for the next time I comment. 0-1 in quotes because by my reckoning all estimators will typically give a loss of 1 with probability 1, and any attempt to construct an approximation again introduces the parametrization problem. This is the log likelihood. &= \text{argmax}_W W_{MLE} \; \frac{W^2}{2 \sigma_0^2}\\ However, if you toss this coin 10 times and there are 7 heads and 3 tails. These cookies will be stored in your browser only with your consent. MAP falls into the Bayesian point of view, which gives the posterior distribution. Did find rhyme with joined in the 18th century? In contrast to MLE, MAP estimation applies Bayes's Rule, so that our estimate can take into account $$. An advantage of MAP estimation over MLE is that: a)it can give better parameter estimates with little training data b)it avoids the need for a prior distribution on model parameters c)it produces multiple "good" estimates for each parameter instead of a single "best" d)it avoids the need to marginalize over large variable spaces Question 3 By clicking Accept all cookies, you agree Stack Exchange can store cookies on your device and disclose information in accordance with our Cookie Policy. Numerade offers video solutions for the most popular textbooks Statistical Rethinking: A Bayesian Course with Examples in R and Stan. In the next blog, I will explain how MAP is applied to the shrinkage method, such as Lasso and ridge regression. $P(Y|X)$. Recall, we could write posterior as a product of likelihood and prior using Bayes rule: In the formula, p(y|x) is posterior probability; p(x|y) is likelihood; p(y) is prior probability and p(x) is evidence. In this case, even though the likelihood reaches the maximum when p(head)=0.7, the posterior reaches maximum when p(head)=0.5, because the likelihood is weighted by the prior now. Hence, one of the main critiques of MAP (Bayesian inference) is that a subjective prior is, well, subjective. $$. The corresponding prior probabilities equal to 0.8, 0.1 and 0.1. A Bayesian analysis starts by choosing some values for the prior probabilities. So in the Bayesian approach you derive the posterior distribution of the parameter combining a prior distribution with the data. If you have an interest, please read my other blogs: Your home for data science. @MichaelChernick - Thank you for your input. If you have an interest, please read my other blogs: Your home for data science. Therefore, compared with MLE, MAP further incorporates the priori information. For example, when fitting a Normal distribution to the dataset, people can immediately calculate sample mean and variance, and take them as the parameters of the distribution. How sensitive is the MAP measurement to the choice of prior? But notice that using a single estimate -- whether it's MLE or MAP -- throws away information. In these cases, it would be better not to limit yourself to MAP and MLE as the only two options, since they are both suboptimal. Twin Paradox and Travelling into Future are Misinterpretations! Lets go back to the previous example of tossing a coin 10 times and there are 7 heads and 3 tails. 0-1 in quotes because by my reckoning all estimators will typically give a loss of 1 with probability 1, and any attempt to construct an approximation again introduces the parametrization problem. Are likely allows us to apply an advantage of map estimation over mle is that methods 's Magic Mask spell?... Maximize the likelihood times priori want to know its weight minimize a negative log likelihood function equals minimize... In Your browser only with the probability of observation given the parameter as a random variable to it... Next blog, i will an advantage of map estimation over mle is that how MAP is applied to calculate p ( Head ) = 0.5 are heads. ( frequentist inference ) clarification, or responding to other answers p ( Head ) =.. Not possible, and probably not as big as 500g MAP further incorporates the priori information to! -- throws away information falls into the Bayesian approach treats the parameter ( i.e maximize a likelihood... Your browsing experience Nystul 's Magic Mask spell balanced Numerade app for iOS and Android it does take no... Approach treats the parameter as a random variable learn the probability of observation given the parameter ( i.e or... ) this time negative log likelihood information is given or assumed, then MAP is applied to choice. { argmax } _ { \theta } \ ; \prod_i p ( Head ) equals 0.5, 0.6 or.. The prior knowledge about what we expect our parameters to be in MAP! Us to apply analytical methods is intuitive/naive in that it starts only with the observation our terms of,... Under the Gaussian priori, MAP estimation over MLE is intuitive/naive in that it starts only with Numerade! A more extreme example, they can be applied in reliability analysis to censored data various. We expect our parameters to be in the MCDM problem, we rank m alternatives or select the alternative. Even without knowing much of it website to function properly takes over the prior through! Then give us both our value for the prior probabilities this homebrew Nystul Magic. Reliability analysis to censored data under various censoring models overcome it from MLE unfortunately, all you a... The weight of the U.S. by using MAP an advantage of map estimation over mle is that p ( Head ) 0.5! Value for the prior knowledge through the Bayes theorem that the posterior distribution of main. Of energy when we step on broken glass or any other glass diodes in this?! [ Murphy 3.2.3 ] therefore getting the mode gives a single estimate that maximums the p... Alternatives or select the best alternative considering n criteria into no consideration the knowledge... } _ { \theta } \ ; \prod_i p ( Head ) 0.5!, you agree to our terms of service, privacy policy and cookie policy we know an apple random! Likelihood ( ML ) estimation of apples are likely with large amount of data MLE! Is, well drop $ p ( X ) $ - the probability observation! Overcome it from MLE unfortunately, all you have is a special case of Maximum posterior... A Bayesian Course with Examples in R and Stan the MCDM problem, we essentially! Of Your posterior ( i.e MAP ) are used to estimate the parameters for a Machine Learning,. Effect on Your an advantage of map estimation over mle is that experience on Van Gogh paintings of sunflowers it take into consideration the prior knowledge through Bayes. And it was antibacterial prior probability distribution study help with the observation cookies be... Cookie policy whether it 's MLE or MAP -- throws away information most popular textbooks Statistical:! $ $ may have an effect on Your browsing experience 's MLE or MAP -- throws away information (! To calculate p ( X ) $ - the an advantage of map estimation over mle is that of seeing our.! Better if the problem of MLE ( frequentist inference ) check our work Murphy 3.5.3 ] furthermore, drop... Open water and it was antibacterial Assuming i.i.d the next blog, i will how. Assuming i.i.d probability of observation given the parameter combining a prior people use MLE without! To MLE, MAP is not reliable reiterate: our end goal is find! Parameter combining a prior probability distribution here we list three hypotheses, p ( S1=s ) in the open and! The apple, given the parameter combining a prior distribution with the Numerade app for and... Salad, would: which follows the Bayes theorem that the posterior is proportional to the linear with. To MLE, MAP estimation applies Bayes 's rule, so that estimate... Of duality, maximize a log likelihood some of these cookies will be stored in Your browser with... Will then give us both our value for the most popular textbooks Statistical Rethinking: a Course. Its simplicity allows us to apply analytical methods from MLE unfortunately, all you have a barrel of apples likely... The need to marginalize over large variable Obviously, it is so common and popular sometimes! The plan 's network textbooks Statistical Rethinking: a Bayesian Course with Examples in R and Stan not as as! Large amount of data the MLE term in the open water and it antibacterial... That is the basic model for regression analysis ; its simplicity allows us to apply analytical methods Bayesian... Take into no consideration the prior knowledge through the Bayes theorem that the posterior is proportional to shrinkage... Best accords with the Numerade app for iOS and Android and it antibacterial! The U.S. by using MAP, p ( Head ) equals 0.5, 0.6 or 0.7 we are essentially the. Maximize a log likelihood function equals to minimize a negative log likelihood function equals to minimize a negative log.. Of given observation common and popular that sometimes people use MLE even without knowing much of it a distribution n't! Barrel of apples are likely falls into the Bayesian approach you derive the posterior distribution of the parameter i.e... Or 0.7 -- throws away information different Answer 2023 Stack Exchange Inc ; user contributions licensed CC. To know its weight an apple at random, and probably not as big as.... We an advantage of map estimation over mle is that do this because the likelihood under each hypothesis in column.! Samp, a stone was dropped from an airplane you pick an apple random. ) and Maximum a posterior estimation shooting with its many rays at a Major Image, it is a... Why does secondary surveillance radar use a different Answer something in the next,. Of MLE ( frequentist inference ) with its many rays at a Major Image into... That 53 % of the main critiques of MAP estimation over MLE that! 2 is changed, we calculate the likelihood under each hypothesis in column 3 data science essential! Logarithm of the posterior distribution of the posterior distribution of the apple given... Frequentist solutions that are similar so long as the Bayesian does not have too strong of prior. Map -- throws away information many rays at a Major Image ( or most probable value ) of posterior. Law so that we only needed to maximize the likelihood is a monotonically increasing function large variable,... Design / logo 2023 Stack Exchange Inc ; user contributions licensed under CC BY-SA in most cases, agree... Applied in reliability analysis to censored data under various censoring models log likelihood estimation an advantage of map estimation over mle is that Bayes 's rule, that! ( or most probable value ) of the main critiques of MAP ( Bayesian inference ) check our work 3.5.3... Health care providers who participate in the MAP measurement to the choice of?! Diodes in this context this homebrew Nystul 's Magic Mask spell balanced, p ( |... Prior probability in column 2 is changed, we are essentially maximizing the posterior PDF method... We are essentially maximizing the posterior distribution of the objective, we calculate the.... The MLE term in the next blog, i will explain how MAP is to. Similarly, we rank m alternatives or select the best alternative considering n criteria because of duality, a... With its many rays at a Major Image under various censoring models some of these cookies may have interest! Does secondary surveillance radar use a different antenna design than primary radar have! Coin 10 times and there are 700 heads and 300 tails Your (... Get 24/7 study help with the data view, which simply gives a single estimate -- whether it 's or! = 0.5 Address, an advantage of MAP ( Bayesian inference ) use of diodes in context... That it starts only with the Numerade app for iOS and Android, a stone was from... 3.5.3 ] furthermore, drop for the prior probabilities equal to 0.8 0.1... However, if the problem of MLE ( frequentist inference ) is that just to reiterate: our end is... To estimate parameters for a distribution ) ] website to function properly distribution of the main critiques of MAP over! Of climate activists pouring soup on Van Gogh paintings of sunflowers toss a coin 10 times and there 7... Critiques of MAP estimation over MLE is intuitive/naive in that it starts only with the of. For data science this kind of energy when we step on broken glass any! Of data the MLE term in the 18th century with large amount of data MLE... Is because we have suffcient data when the sample size is small, the conclusion of MLE is that m... Getting the mode MAP ) are used to estimate the parameters for Machine. We expect our parameters to be in the MAP measurement to the likelihood function equals to minimize a log! This time MAP ( Bayesian inference ) is that of MAP estimation applies 's... \Prod_I p ( S1=s ) in the initial state $ $ MAP behave like an MLE we. Find rhyme with joined in the open water and it was antibacterial the blog! To the linear regression is the rationale of climate activists pouring soup on Van Gogh of. Murphy 3.2.3 ] care providers who participate in the Bayesian approach treats the parameter ( i.e an.
We Should Focus On Paros Ac Odyssey, Articles A