Maths Key Skills Stage 6 Skill Check 8 Answers: Fill & Download for Free

Firstly, I’m going to assume that we’re on the planet Mercury where a week is roughly 58 weeks. Assuming you spend about 8 hours a day learning for 5 days a week, you have 2320 hours, enough to have a good enough grasp over deep learning to pass off as a research engineer, and a beginner level researcher.So first off, do yourself a favor and get hold of the Deep Learning book by Goodfellow et al. It’s an extremely SOLID book on deep learning that doesn’t shy away from any of the math, while not being overly pedantic. For example, it has a great way of explaining what happens when you optimize KL divergence the other way, what L1 and L2 losses actually do to your weights (in terms of eigenvectors of the loss function), intuition behind LSTMs and so on. Start with the first section of the book. It’ll take approximately 2 weeks to completely digest that material (YMMV). If you’re past this week without getting bored or bogged down by the mathematical complexity, you’re now ready to graduate to some machine learning.At this point, we need to stop reading, and start coding. But to actually understand stuff, do not download Tensorflow, start with implementing basic things (Logistic regression, SVMs, kNN, PCA, GMMs) in good ol’ python. A lot of people just rush to using high level libraries without understanding how to code up basic blocks from scratch, and you’ll learn a lot if you do so. For example, learning that you need to standardize your data before applying PCA, how to make softmax numerically stable, why you shouldn’t reuse your softmax function to implement log-softmax and so on. To get started with implementing these, head over to CS231n and finish the first Assignment, building up on that to implement SVMs and GMMs. This would take approximately 4 weeks (assuming you start off with no knowledge of numpy or python) to implement and debug. Now you know how to implement basic machine learning algorithms from scratch.You’re now good to go ahead and read the excellent notes on the CS231n website (till Module-1). Take your own sweet time to digest the intricacies of optimising non-convex functions, and yes, IMPLEMENT every thing you read. From momentum based methods to basic backprop, implementation is the key if you want to actually learn the mechanics. For example, if you implement backprop on your own, you’ll realise how you need to keep a ton of data around for the backward pass and this knowledge could be very helpful if you encounter out of memory errors at some stage in your DL engineering career. This will take another 4 weeks.It’s been 10 weeks of insane DL now. You’re now solid on your basic math skills, and can implement feed forward nets and basic ML algorithms in raw python. Congrats on that. Let’s step it up a notch from here on. It’s only going to get harder (and thus more fun).Time to take out our Deep learning book, and start from where we left off. Read chapters 6, 7 and 8. Chapter 6 is easy stuff, and we’ve already mastered that. Chapter 7 is where it starts to get hard - but the book does an excellent job at distilling math heavy concepts into easy to understand text. To supplement your learning, do watch this amazing lecture - Deep Learning Lecture 11: Max-margin learning, transfer and memory networks.This will take another 4 weeks but with this done, we’re ready to graduate to CNNs and RNNs. We’ll follow the same philosophy again - read notes from cs231n, implement and train dat net, and then read the book to understand the finer details. This phase will and should take longer because implementing backprop in CNNs (with all the bells and whistles such as batch normalization) isn’t all that easy. Same goes for LSTMs (watch the norm of that gradient, and clip if it gets too large, or truncate to shorter sequences if it gets too small). Pro tip - for training LSTMs, I do recommend checking out the PennTree Bank dataset and train a language model to get a perplexity of around a 110. If you can make that 90ish, congrats, you’re doing it right. To train a CNN, just do assignment 2 from CS231n.Phew. that last phase of training CNNs/RNNs and then reading associated material from the book, set us back by about 6 weeks and we now have just 38 weeks in our pursuit to become the very best.Finally, time to pip install tensorflow. Do not even think of anything else. No, we’re not doing pytorch (it’s dynamic in nature so not great if you want to get a hang of autodiff software, but great otherwise) or theano (not large enough). Or high level libraries like keras (beautiful, but makes it all look too easy). Head over to the docs and read the MNIST tutorial (for experts, coz that’s what we are now). It’s all going to be very weird at first, the idea of a computation graph, sessions, and the fact that there’s a separate “compile” and “run” even within python itself. But once you’ve got a hang of things, you’ll begin to realise how beautifully it’s been designed and how easy it is to extend - for example, writing your own LSTM is just as easy as extending a base LSTM class. As a set of warmup exercises, do read the code in the tutorials (Tutorials | TensorFlow), and step through each tutorial very very carefully, understanding what each line does and why. You’ll learn how to organize your tensorflow code. Pro tip - I always like to have separate abstractions for training, testing and predicting. And a separate build() function that builds the entire computation graph, starting from placeholders, all the way to defining the loss function and the train_op that optimizes that loss function.Shameless plug - I recently coded up a model for machine comprehension (a task I’ve grown to really like). To see how to assemble your model code together, do check out MurtyShikhar/Question-Answering. To test your tensorflow skills, translate your CNN and LSTM code to tensorflow to achieve the same (maybe better) accuracy/perplexity.Learning tensorflow took about 8 weeks of insane hacking, reading tutorials, getting frustrated over dem tensorflow compile errors but I hope the journey was rewarding. You now have the power of autodiff and a really cool arsenal of functions to wire up literally ANY deep learning model ever. Before we move on to some research-y things, why not try out a fun project in tensorflow? Pick up any cool paper on a well defined problem (Machine Translation/Image Captioning/Comprehension/Question-Answering/Visual Question Answering/Style Transfer) and implement the state of the art model in Tensorflow. Do not try GANs as yet, coz we ain’t that cool (yet). You’ll take about 5 weeks to finish off your project and get competitive results leaving about 5 weeks for just reading about current research.I realise that 5 weeks are really a short span to cover the breadth of DL research, so why not narrow down our scope? Pick up one problem or technique - be it Generative models, or seq2seq models or reinforcement learning. I don’t know anything about reinforcement learning (I’m trying to learn though), so I’ll focus on the other two.For Generative models, start with this amazing tutorial by our lord and saviour - NIPS 2016 - Generative Adversarial Networks - Ian Goodfellow, and slowly follow through, pausing and googling as and when required. You’ll find those math skills from the first few weeks VERY useful here. I then urge you to read the original GAN paper , and followups such as DCGANs, SeqGANs, and the newest and coolest WGAN.For Seq2Seq, watch this amazing lecture by a Google Brain scientist - Sequence to Sequence Deep Learning (Quoc Le, Google) and then check the ICML 2017 tutorial by Oriol Vinyals - seq2seq icml tutorial.The END.That took some time and effort. It is the end of week 58 and you now have a broad understanding of Deep Learning, along with some solid math and engineering skills. It is time to venture deeper (get the pun?) into Deep Learning research, or maybe build a cool app. I leave that decision upto you, you stud DL engineer.

Memory power in relation to your study is a matter of following proper study technique.Study TechniquesHow much to study in a day1.Maximum quality input cannot be more than 7-8 hours. More than this will be waste of time. You will fail to absorb and therefore feel bad that you are a dull student.2.Keep all other subject books, except the one that you study, out of your sight. Avoid unnecessary mental burden. It will help you concentrate and improve efficiency.3.Set practical targets for yourself. Impossible targets are depressing. If time is less and pending work is more, then the best would be to decide what to leave.4.Have proper sleep and rest. You cannot add to your study hours by sleeping less.5.Mark days and hours for each part of the syllabus to be covered and follow it religiously. Keep making minor adjustments.6.Keep some time aside for breaks and entertainment. It will not be waste of time but will increase your output.7.Morning hours are most productive. Study subjects like Math at this hour.Study Techniques1.Do not use help books straight away. At least first two readings should be from the text book.2.Read with a soft pencil in hand. Mark or underline any important points. After reading a topic or a sub topic mentally review what you just read. Do not go on mechanically reading without understanding. Otherwise when you read it the second time, after a few days, you will be as blank as you were the first time and that will frustrate you and hamper your efficiency. After you are sure that you have understood what you just read, note down its summary on the page margins itself.3.The summary should be such that on reading it the whole of the matter comes to your mind.4.After you finish a chapter try to attempt questions based on that chapter. You will be surprised to discover how much you missed. Read again so that you are able to answer the questions properly. At this point you may consult a help book to check your answer but only after attempting the answer yourself.5.Attempt all types of questions; objective, multiple choice, short answer and long answer. Take questions from previous years’ papers. These are in fact the best source. If the topic has a diagram, draw a neat diagram yourself and label it and understand the process from it.6.Carefully remember formulae and equations.REVISE, REVISE AND AGAIN REVISE. AND FINALLY REVISEIt has been scientifically established that one tends to forget three fourth within a week of the reading. Memory graph is a steeply downward sloping line. What should one do to overcome this? Studies also show that if you revise a matter within hours, the retention improves significantly. Ideally you should revise a topic next day and again at the weekend. Revise again in a month.How to find time for so many revisions? Read a topic carefully twice underlining the key phrases. That is one revision. Second reading will take less time and will be more productive. Jot down a summary in page margin. It will be second revision. Solve questions based on the topic. It will be third revision. On the weekend revise just the summary or the key points. It will be fourth quick revision. Finally do a monthly revision. When it is exam time you will not panic. You would have remembered broader points. Quickly revise the whole syllabus paying special attention to some tough topics/questions/formulae/equations.Mind MappingMind remembers diagrams and colors more easily than the text and numerals. Jot down the key points in colored pens. Encircle key words in pink, blue, red, green, yellow etc. Use different thickness pens of different brands to make a different imprint in your memory. Make a group of key words of a topic together. When you read those again and again during revisions, the colorful images of the words will etch in your memory. Make diagrams of the words even when there is no diagram in the question. Your mind will easily bring them to front when you need those at the time of writing answers.Practice, Practice and more PracticeIf you find a particular subject or topic difficult, learn its easier part first. Leave out the tough part initially. If there is a tough question, read a solved example. (There are solved examples in most of the Math books. You may do it from help books too). Solve a question again and again till you feel at home. A stage may come when you will be surprised as to why did you feel it so difficult to begin with.Relax – Take 20 Deep BreathsIt has been scientifically proved that deep breathing relaxes mind, settles down fast heart beat, improves concentration, and improves efficiency. The art that began in India has now been adopted all around the World. It costs nothing. It takes very little time. Most of the students are under great stress when exams are near. So relax. Take 20 deep breaths three times a day like you take medicine. Believe me you will be surprised at the positive impact.

Teaching can be boring. Learning is best when you are having fun & don't realise it is happening.Some activities:Skill: Pouring milk into a cup.On a sunny day, take 2-4 yr outside, with 2 liter/quart milk bottle filled with water. Start with it partially full. Have them pour it into plastic cups. Initially it will go everywhere. But it is safe to fail. Eventually your 5-6 yr can pour from a full bottle over carpet without the risk of a disaster. To make more interesting. Pour into glass containers, fill to different heights & tap for different notes.Skill: Counting.Just before bed. Let them jump on the bed while you hold their hands & count. Transition to them counting their jumps. Transition to they can jump for as many as they can correctly count.Skill: Reading.Mix it up. Read to them a more complex story. 1 chapter a night. Also have them read a simple book to you.Skill: Strategy & motor skills. pure fun. Play stompy feet.The goal is to have one player put their feet on top of the other players foot to get a point. First to five wins. You have to move your feet out of the way, while at the same time getting in position to strike. You play in bare feet, near bed time.As a parent. Keep the scores tight. Be within 1 point in front or behind. Mostly let them win. If you are playing best out of 3, you can win the second game, if the score was 4:4 for a while.Objective: Fun time with dad.Age: 3-8-ish.Be careful about letting siblings compete against each other. They are used to competing 100% vs dad & may hurt each other.Skill: Road safety. Age: 3-18.When you go out, put a child in charge of saying when the family should cross the road. They like the responsibility & trust.Tell they if they say go & their is a car coming all the family will walk out & get squashed. So you need them to look carefully & do how they've been taught. If they do say to walk & it is not safe, stall by asking a question "ie: Did you remember to look for cars coming around the corner?" By then they may see the issue & say the family should wait some more. Never veto their decision, just delay & have them change their mind.NB: "Don't walk" means "Don't walk". But "Walk" means "Check for traffic & if it is OK, then Walk. And as you walk keep checking as some drivers run red lights"Skill: Motor skills & bonding. For 2-7yrs.Take them to a park (or a Macdonalds) which has climbing castles, slides, etc. Do it regularly. Say each Sun for 1-2 hrs. Mix it up & go to different venues. They just get into the habit of doing something fun with dad & mum gets a rest. For sporty kids you can mix in throwing frisbee or balls. Don't force it, only one of my kids really loved ball sports. The rest got bored very quickly.If you live near a beach build sand castles or dig holes & let the water seap in.Skill: Family TimeAs kids grow. Evolve "going to the park" into regular Saturday night "board game night or cards". Something they enjoy but you need to interact with each other. If necessary it can be compete against each other on a computer game. (This is not as effective but might work if you are starting with older kids)Also good is a regular Sunday BBQ lunch with the extended family (or families of their friends if no relies nearby)Skill: EducationDon't give your toddlers access to games on your touch screen. Their brain is still developing spatial skills, fine motor skills & 3D awareness. The 2D world of touch screens slows this skill development.Buy some good educational titles. (Maths, English). Games that don't just ask questions but actually do things with puzzles you need to solve. eg: Asteroids fly at your ship with sums on them, if you type the correct answer in time it blows up. Else your ship gets hit.Done properly, your kids will start school with skills of those in year 3 maths, science & english.KEY: Sit them on your lap & be with them while they play. That way you can help them & it is fun to play with dad. Rarely will they play for long by themselves.KEY 2: They will quickly lose interest in these games once they are exposed to regular games (ie: Halo, COD etc). Try to delay it as long as possible.Direct educationThese fun times often result in quiet times, where the kid has "run themselves out". So they are happy to sit in your lap & cuddle. or sit next to you eating an ice-cream & talk.Often you will both see something that you can use as a catalyst for a lesson.Conversation: "Take nothing but photos, leave nothing but foot prints"eg: Broken Glass or litter in their playground. "Isn't this annoying they were so inconsiderate to wreck our fun? Lets clean it up.Conversation: TrustTell them that one day they may get into trouble & they will really need you to believe them & to trust them. So the most important thing is that they are honest with you. It doesn't matter whatever bad thing they do they absolute worst one is to lie to your dad (or mum)Note: You need to set the example by being completely honest with them. ALWAYS.Conversation: "When you are sick some drugs can make you better, When you are better all drugs can make you sick". Age: 2 - 30Tell them they have your permission to drink, smoke &/or take drugs. It is their decision. Your only requirement is that if they decide to do it, they must tell you & not hide it.That said, take every opportunity to expose them or educate on the downsides of that choice. eg: watch documentary, listen to smokers coughing their lungs out. (but be discrete). Take them to visit the cancer ward in your local hospital. 90% of the patents in there are smokers with lung cancer.If in public point out kids spewing or hang overs & highlight their body doesn't like alcohol & that is why they feel sick.Note: Don't attempt to teach "responsible drinking" by giving them a few drinks at home. Many studies show the younger the child acquires a taste for alcohol, the more likely they are to binge drink or develop problems with alcohol as they get older.Conversations: Sex Ed.Start early: If you talk to you child about intimate subjects like puberty, periods, STI's etc, when they are too young to be embarrassed. ie 8-9 then you can continue to have an influence in important intimate topics thru their teenage years. ie: "it is possible to get p regent the very first time", Ensuring they are on contraception. Sex changes your brain chemistry which distorts how you feel about that person. That can be great if you have a child together, But can be devastating if you start too early, they dump you & you are stuck at the same school with them every day.Most Important.Your children learn most by watching & imitating you. If you deceive them with "white lies", shout at them or your spouse, if you get angry & hit them. Get drunk, rage abuse toward people while driving or at other times. Smoke cigarettes or other drugs. or otherwise set a poor example, do not be surprised when your teenager ignores, lies & shouts at you.I repeat. Your kids will not "Do what you tell them" if your actions contradict your advice. If fact they will stop listening to your advice too.