I wish to pivot from business intelligence to immunoinformatics but there aren't opportunites outside USA :(

So excited to see your videos again!

1:10:10 - Conditioning and Stability 1:13:38 - Approximation Accuracy 1:16:15 - Expensive Errors 1:19:06 - Memory Use - Sparse vs Dense 1:21:20 - Speed - Computation Complexity 1:23: 20 - Speed - Vectorization 1:25: 40 - Speed - Locality 1:39: 36 - Speed - Temporaries

I am so thank full that I found this course. Alhumdulillah!

Starting now the nlp's studies. Great video

The host has such an annoying voice,bruh

Great video! Thanks!

Whats your take on ketogentic diets on gut health (e.g. carnivore and keto)? Changing the energy (fat or carbs) entering your gut is going to affect the type bacteria and fungus flourish (e.g. ones that like fat are going to increase and ones that like carbs are going to decrease) but im not sure on the evidence of it being positive or negative for the things mentioned in the video...

fascinating

Welcome back Please keep em coming 😊

hey guys now it is worth it to do this course or it is outmoded

When I heard "topdown approach" at the start of the video, I was convinced! Thank you so much!

Kak Rachel, saat ini tanggal 21 februari 2024, saya ada pertanyaan buatmu, kenapa AI yg ada saat ini hanya memiliki kemampuan untuk menjawab pertanyaan ? Mereka tidak bisa bertanya kepada pemakainya. Apakah AI bisa dilatih untuk berinisiatif bertanya ?

This is what I have been looking for. Thank you

I do not know why but I hate your voice tone.. hope you can change it so I can follow your videos ..

24:39 Why we transpose both matrices ?

@16:45 : actually the Hilbert matrix is symmetric and positive definite. So in theory Cholesky works - numerically it is a different story

46:30

Thanks for the great course. NMF using scikit learn should be like this: clf = decomposition.NMF(n_components = d, random_state=1) W1 = clf.fit_transform(np.array(vectors).T) H1 = clf.components_ show_topics(W1.T) vector.T is matrix of (words, doc) while vector is (doc, words) and it does not make sense to decompose vector.

How Grad of R = W@H-M is [ [email protected] W.T@R ] ???

Good Video about SVD as well as you talked about Gilbert Strang which was even better. Decomposition of a matrix into three matrices orthogonal columns, diagonal, and orthogonal rows matrices. Good one!!!

the example of GRU was horrible, I had no idea of GRU till i saw another video about this topic and realized how much simple it is

the encoder mask is None?no mask for Encoder

When someone does their work enjoying it, the outcome always comes to be the best. Thank you Rachel :)

Thank you

maybe women and people of a diverse bent aren't that into ML

📺💬 What is the number of patience in each state at year 1 🥺💬 Start from the state 1 Asymptomatic with the number of patterns assume 100 patience after first yeat 1st state is 100 * 0.90, 100 * 0.07, 100 * 0.02, and 100 * 0.01. 📺💬 It is the Markov relationship that explains each state and the probabilities of their relationship. 📺💬 Starting here at first state is 85 percents, 10 percents, 5 percents, and 0 percents. 🐑💬 With 85 percent number of the patients will be 100 * 0.09 of 85 percents 🥺💬 The 10 percents of state 2 will be 100 * 0.93 + 100 * 0.07 of 10 percents 🐐💬 See from the Matrix it is the current row times relationship adding with the previous state. { 76.5, 15.25, 6.45, 1.80 } 🧸💬 result: tf.Tensor( [[50. 49. ] [58.5 61. ] [43.5 43.5]], shape=(3, 2), dtype=float32) 🐑💬 The best with value of price is the lowest of Dot product for S1 and S2 for each P1, P2 and P3 📺💬 The reason we divided imge with .255 or 255 it is the properties. 🥺💬 The input can be dtype = float32 any number can multiply by the target but some pictures converted from png or jpeg format they are in unit8 or other format using image_to_array or make it as float32 by multiplying by .255 or dividing it by 255 is make senses when it is colors pictures that make the input sample in the rage of [ 0 to 255 ] with dtype = tf.float32 📺💬 We trying to make it between [ 0 and 1 ] 🥺💬 That will accept both [ 0 to 255 ] or [ 0 to 1 ] but float32 format and no out-of-range value including in the matrix value that will use the default ranges. 📺💬 You can multiply or divide as long as it is in the target ranges. 📺💬 It is the sample filters, TOP and it is not the correlations when they explain of S1 and S2 or rows of data but think it is contrast number that make the edges see significants. 🧸💬 What is it about you times the picture matrix with the correlation matrix ⁉️ 🐑💬 That is the idea when you do it with the convolution layer but the initial value and weight results matrix is learning over time. The result you can extracting the value of the convolution layer weights. 📺💬 It is possible to divide image by image ⁉️ 📺💬 it is possible but the purpose is to create float32 value with specific ranges and trying to make the contrast image from another image you need to select the specific image from some algorithms such histograms or blurs pictures. 📺💬 Exercises 🥺💬 It starts looking like the sollinoids coils or running tubes. 📺💬 A curves of cyclones. 📺💬 You can try making it as a unit it repeats with same patterns until repeating itself as one. 📺💬 Double-precision when you need to stored all information into a length of bytes of in telecommunications you can think it as SI units by divided them by the target precisions and again and again until you have target precision. 🐑💬 Imagines how you to send those of high precisions over networks with laggy with long bits length repeating of number possibilities. 🧸💬 I agreed the idea but noone use the IEEE precisions except the database or specific telecommunication programs or GSM. 📺💬 There it is the Type 2 where Python in general 📺💬 what kinds of information they are using⁉️ 🐐💬 There are many but what is easiest to understanding is the measurements or locations where you missed of a bits of transmission you still current locations. 📺💬 There is something called the machine epsilons representing for computers 🧸💬 Correct when more precision is needed but communication is to send all the information in the same piece you request and understand the same that is what GSM does. 📺💬 Do Tensorflow has something as the Eigenvalue ⁉️ 🐑💬 Yes it is . 📺💬 The example of blooms filters 🧸💬 That is very nice example when you can create surveys and import direct datagram using panda, I see what is the target of Bloom filters. 🧸💬 False positive, someone tries to make mistakes by attention such as input address/hyperlink or multiple tabs or intention response the answer. 📺💬 Intel microprocessor, the floating points claims 📺💬 Do you have any of problems about Intel processes ⁉️ 🐑💬 Yes but his project is very very very tiny and small compared to that when hibernated and power steps were introduced that make those telecommunication programs need to learn about intel internal clocks and OS clocks. 🥺💬 I will be right back almost 18:00 and I views read of this lesson again. 📺💬 The area located where you can work with the algorithms of speed and area coverages. 🐑💬 How much of the area are effects by speeds and causes some feedback actions over some value⁉️ 📺💬 Locality is how much of data in sequences presents of similarity ( locality and similarity are different when locality represent existing or described patterns but similarity is what the value properties are ) 📺💬 Choices space when time running with the same actions possibilities outcomes increase. 📺💬 Example of scalability and parallelism 🐑💬 It is when you can scales of the problems you can perform better with tasks such number of students who came to class and reply the question with attention or homeworks assignments. 📺💬 I like your jokes we starts from an assessments 🧸💬 😠 ...

📺💬 Does Anyone knows what the BatchNorm actually does ⁉️ 🧸💬 it is the normalized batch that creates some space for learning when groups of data have shared initial properties. 📺💬 It is normalized of the activation prevention from it goes to zeros or overflows 📺💬 It is not the exact value but it is true when significants are close and the activation functions determine of consideration point between them. 📺💬 We do not need it anymore, have new constants at the end. 🥺💬 I am asking for the application on target machines not to copy all the model output but all the works from neurons and equations can be extracted into one-dimension arrays that robots understand. 📺💬 Elementwise multiplication when sorted of the columns while working solved the multiplication. 👧💬 Sorted inputs, that also exiting in the LSTM and that is what the logical gate doing. 🐑💬 Correct that is the natural source of sorting order. 📺💬 That is what you are formulating running the result start to sorted and groups that are simple steps what the neurons learning. 🐐💬 You can try with Piano key notes when you rest from working it starts by indexes of some keys then multiply until sorted but how you make the label. 🧸💬 That still required some guiding even un-supervised learning. 📺💬 Listen for the unsupervised learning you may try to calculate and keep in your long terms memory. 🥺💬 Absolutely there must be someway identified the significant methods to the results. 📺💬 We are learning about the GRU but I wonder if there is no DropOut in our model. 📺💬 We can add it next time but how about LSTM they are learning from addition and forgetting gates. 🧸💬 He may remember from some applications they using GRU leans the values the expansions and reduced is the idea but leaves significance for the rest.

📺💬 Can we remind you the Bleu score is ⁉️ 🥺💬 The first time I watch this VDO and I think the Bleu score is a comparison of models or transformers by its definition that could bring some advantages when models or transformers can use to answer the questions. 📺💬 Specifically the n_grams models can work with 2_gams, 3_grams, and 4_grams for the suitable components. 🧸💬 Is that will help with pronunciation problems when 2_grams and 3_grams are used I had a bit of reading on these topics. 📺💬 Yes, our example is the Uni-grams or Bi-grams but it is not the best we are learning to match more suitable methods for the questions. 📺💬 One of my example on the Belu scores is that they compare the recognition rates between he ate deer or he ate and apple that the models are suitable for the questions. 📺💬 Yesterday I tried to make something Elvis with the mass MCSS, transformers, and transfer learning ... 🥺💬 That is interesting when the transformer is learning when transforming the information or communication encoder/decoder and transfer learning is something to understand about how they are learning. What are you specifically try on the experiments ⁉️ 📺💬 MASS - mass from Microsoft research. 📺💬 Does Anyone remember what the teacher forcing is⁉️ 🧸💬 That is more influent of supervised learning when we conduct students as the learning neurons and teacher give them advice on specific usages as you are doing now we are listening and answers when you questions. 📺💬 Anyone remembers the Sortish like⁉️ 🥺💬 By it the definition it creates matrixes representing the input as the properties of the sorted sequences in results when the padding creates better much of information but the purpose is to have the outcomes from alignments and matching methods. 👧💬 It is can write into anything as long as it remains the same properties. 📺💬 That is something when you have a kind of specific lengths to it you may need for someone to make it short to make it possible to understand by user. 📺💬 How about we represent sentences into a single one-dimensional array that can save a lot of processors and labels. 🥺💬 It is possible when labelings with the features output of the transformers you had encoding/decoding, She introduced you to the padding methods and that is also the input of the transformer. 📺💬 There is a transformer without the padding 🥺💬 Yes it is that is because of the properties of the transformer, they are studying the neurons networks that is because to enchants the abilities of the transformers. 🐑💬 The difference is transformers are specifically transforming input and output but neurons are combined learning with multiple layers and methods that is why neurons can consider multiple methods input at the same time. 🥺💬 I am reading to reviews about my learned lessons, GRU is the Gate Recurrent Units the definition is Logical gate working on specific value and it is a very good performance on sequences input that had repeating values but it will use a lot of performance try using Bi-Direction Layer with the GRU that will cost a lot of processes ( I running on my old Desktop since my study in school ) 🐐💬 More problem of GRU is it does not provide clues for the next layer that may not be required for specific tasks but relatively is important in none relative data. 📺💬 There are some differences between the LSTM and GRU when the GRU is specifically used for some process and when the sequence is longer then it is hard to hold all that memory and required a lot of processes. The LSTM and GRU is only mini of different in the architecture. 📺💬 Does Anyone knows about the REST BLOCKS and REST NET residual networks ⁉️ 📺💬 Some of the input is the image and using differentiation to determine the significance. 👧💬 Do you mean the concatenated model provides better results or they are specific tasks at each layer ⁉️ 🐑💬 It is possible to consider at each corners of the objects and concatenate the results. 📺💬 I see the concept of the MatMul for the result of the layer but here they use Math.Sqrt what is the concept of this layer ⁉️ 📺💬 It can be anything including Additional, Subtraction, Multiplication, Square, Exponential or power but the embedding layer is has a nature of sequences that respond to squares fn that is all you try print output from this layer. 📺💬 Do you mean the tri-angle of the language encoding/decoding ⁉️ 📺💬 Yes when we predict the sequences with a specific prediction it can reflect of the direct result and back propagation when it may not have meaning but some AI use it to train with another language which is the tri-angle of the language encoding. 👧💬 Do you mean when speaking we also send out our identity and can be generated of new sequences. 🐐💬 More close examples we can have decoding text from input text both of decoding and none decoding are indicated of logical transform yes with encryption sometimes you need to fixed the message or filling it for translation meaning that is what you added your identity. 🥺💬 Don't mind my 5 minutes time break the message from advertising they also leave the Key and Values I can generate a new sample but it is designed to have two times two-dimensional choices to create meaning. 🐑💬 The background we are Telecommunication students and there is more messages. 🐑💬 From the advertising they leave this { H, E, L, L, O } 👉 Foward { N, K, R, R, U } 🐐💬 Advertising 🧸💬 News, HAVE ORCA MUKA SEE EX QC IT JACKS CROSS THNXS LEOERDOE SHOP‼️ ( 🐨💬 HACKS SENSES LESS SINCE EX QC THEY KNOWS NO TRACKING EVIDENCE OR TARGET IT QUICKS ) VOLA COKKIE RUN! =========================================================================== GRUB KOQUUI XAT' =========================================================================== IBCF PTPFBP ELS. 👧💬 Comeback‼️ BirdNest Hidden anywhere else.

Can you please update your code that works well with libraries of today's version? e,g. spacy lemmatizer do not works in spacy v3. whereas spacy version 2 is no more available

Exactly what I would expect from a Swarthmore grad! Great job!

Thank you for sharing this short talk today! I let it play in the background this morning and enjoyed it.

Good explanation but the text appearance is poor

Do you need to have done a undergraduate level linear algebra course for this? I have only studied high school level linear algebra. Will that work?

*from spacy.lemmatizer import Lemmatizer* gives the error: *ModuleNotFoundError: No module named* 'spacy.lemmatizer'. Do you know the reason of thi problem? Is there something I can do?

The whole analogy of Car companies and Tech companies has put Rachel into difficult position because this analogy has flaws. My behaviour at restaurant is under the surveillance of restaurant management and if i choose takeaway service then restaurant is not responsible for my action while eating. Car companies provide takeaway services and Tech company works as a onsite type services.

Great video Rachel, thanks! Wish Goodhart's/Campbell's laws were referenced by name when I first came across the ideas in 2020! Subscribed.

Thanks. Then is there any reason to use regular PCA at all?

Great explanation, thanks teacher Rachel. Good thing that the microphone box is really padded and soft haha.

Thank You🌹There Is Value In Your Work❤️

I'm not sure exactly what they mean by conditional likelihood, but it seems to me that they're probably calculating it incorrectly. "Then, given the knowledge that a review is classified as positive, the conditional likelihood that a token $t$ will appear in the review is L(t | +) = C_t^+ / N^+" Because a term can appear multiple times in a document, the resulting likelihoods can be greater than 1. Rather than using a count of the number of times a term appears across all docs, shouldn't we be using a count of how many documents the term appears in? Assuming that there is some good reason for defining L0, L1 this way, I don't understand the statement "likelihoods are proportional to probabilities." in 9D.2. Why is this true? What exactly is the factor by which you can divide the likelihoods to get the probabilities?

I've really tried to convince myself of this over the past 6 years, but as I'm about to fail calculus again in college, I think I've seen plenty to know this is a huge lie. What's pushed me over the edge, is the kid next to me. While I show up early to each class and take all the notes I possibly can, this guy that sits next to me walks in 10 minutes late every single day, takes almost 0 notes, and doesn't come to class prepared. Despite this, he is almost acing every quiz, while my scores, despite hours of preparation, are all between 3 and 5...out of 15. Dude barely shows up to class, yet he's got an A while I'm failing. I have no other explanation other than he's just innately better at math.

thanks but no thanks because your content is very 'eurocentric'.

Is there any plan to update this course? Thanks in advance.

i hate your videos do better ur not even cute

I am from India and your this course is suggested by many biggest ML or IT youtubers in India . To create Base for ML and AI. you're super cool teacher .

Covid must be an absolute panacea for you AI UN & wef ethical ones, you will own a lot and be happy