Presentation for the Introductory topic of Algorithms

Editor's Notes

• #2: Alan Winfield of Bristol Robotics Laboratory built an experiment where a robot had to save another robot (a proxy for a human) from falling into a hole. Became more interesting when there were two robots to save: robot sometimes saved one, sometimes even saved both; 14 out of 33 times though, it had decision paralysis. Robots follow instructions but does not know why (or the ethics behind it). Imbuing a robot with ethics is a hard problem: important especially with the advent of self-driving cars. Algorithms for military combat robots could be a good starting point – there, robots have an ‘ethical governor’ (a set of algorithms) that helps them make smart decisions in the battlefield. Drones with this programming have been shown to not shoot (or to reduce casualties) in areas protected from combat (like schools and hospitals). Bottom line: We need to know what assumptions we are making when designing the algorithms – the military robots are one field of application where the rules are ‘known’ so it is relatively straightforward to make ‘ethical algorithms’ for them. The interesting part is in trying to get robots to behave well in circumstances that neither it nor we have encountered.
• #3: Computers are only 50 to 60 years old. Algorithms have been around for thousands and thousands of years. The word itself was seen (in its current form) in the 1690s, though in other variants since the 13th century.
• #4: Techniques to make a basket, for a ritual ceremony, for the Macarena! Multiplication in elementary school. Babylonians had algorithms to find square roots and to do factorization. Euclid’s algorithm to find the GCD of two numbers is considered to be the oldest algorithm still used today (300 BC). Genes encode proteins: there are start and stop locations (“codons”); there are step-by-step processes to transcribe these genes and produce proteins; these proteins fold and interact with each other, and the environment, in particular ways – these result in life and living processes.
• #5: Don’t need to be computer based.
• #6: Length of a word: What if you didn’t have a block for that? Interact with the user: Input/output (I/O) technique Word comparisons: Comparing a secret word with a guessed word until the player has found it Make this a block: Wrap things in functions. Don’t forget: Clicking something triggers an algorithm…
• #7: Lunh algorithm: used to determine if a particular number is a possibly valid number. It adds a ‘checksum’ digit (a redundant digit) that is appended to the number; if the checksum does not correspond to the rest of the number, then the number is incorrect. Netflix had a contest to improve suggested movies, the prize: One Million Dollars PageRank works by counting the number and quality of links to a page to determine a rough estimate of how important the website is. The underlying assumption is that more important websites are likely to receive more links from other websites EdgeRank algorithm, which used three different factors, has given way to an algorithm that reportedly uses more than 100,000 factors (but is proprietary).
• #8: Sequencing: When you saw your first program, which blocks go before which? Which steps happen before which steps? Selection: Which part of the code gets executed based on a condition? Forks in a road? If/else? Iteration: Do something over and over until a condition is met. Recursion: You’ll see this later in much more detail. To define something, you use itself on smaller inputs. (Example: Eating chocolate.)
• #9: Putting together two algorithms to complete two tasks produces an algorithm that now completes both of those tasks. Smaller part of algorithm that does a certain task is also an algorithm that now does a sub-part of a bigger task. There are many ways to do something – examples: Guessing the number from 1 to 100 sequentially or using hints like too small, to large sorting a list using bucket sort is quicker than Bogosort Too look for possibilities of 1 billion friends on Facebook could take 30 years or 30 minutes
• #10: Tried to determine what happened after the big bang First to propose very expensive computers that run very fast Arizona has BIO5 that runs experiments on DNA generating terabytes of data This comes back to our computer science department to look for patterns, in the hope to do things like cure, o prevent cancer.
• #11: Algorithms should be correct – this is more important than being fast and efficient. In this class, we will not be too concerned with the efficiency of your algorithm, so long as it is always correct. We might ask you how efficient your algorithm is. TOTAL Correctness: Always returns a value. This is the best scenario for an algorithm (and if it is the most efficient, even better). Examples: Luhn Algorithm, Mathematical Calculations, Encryption and Decryption PARTIAL Correctness: Perfect Numbers (28 is one), the “halting problem”, the “is there a bug in my program?” problem PROBABALISTIC: Getting the exact answer takes a really long time: Traveling Salesman Problem; weather predictions; file download times; existence of a number in a huge, huge list (or the mean); problems where even your input is probabilistic (search queries).
• #12: “Lather, rinse, repeat” joke. We have yet to reach the goal of programming a robot by speaking to it. NLP (natural language processing) is a hard field because of all the ambiguities in language that computers have yet to be able to understand very well. Context is also very important in natural language (big deal when Google and Bing came out with contextual search.) In the case of the robot with the ethical dilemma, unambiguity is tricky because we may not know of all the possible inputs or contexts. That’s why the output might not always be what you expect – this is another example of a probabilistic algorithm. You truly understand an algorithm when you are able to program a computer to do it – a computer merely follows instructions. The computer is never wrong; the output might not be what you expect, but that’s only because you provided an incorrect or an incomplete program to the computer.
• #13: Image represents the different ways of specifying an algorithm to do something. Natural Language: Describe your algorithm to me in a way that I can understand and implement it. In fact, one way of refining your algorithm is to talk it out with someone -- through talking, you get to understand the assumptions that you are making that the other person may not make; through talking, you might also stumble through the answer because you are able to tease apart your thoughts. Also, ‘rubber duck debugging’. Pseudo Code: Semi-English; mixed between English and programming language; “code” that is similar to what you might write or drag in your program, but not bound by any language or any syntax is particular. Uses constructs common across languages, such as if-conditionals, while-loops. Can sweep entire assumptions under the rug: “Find the length of a word.” Flowchart: Pictoral representation of an algorithm – squares represent statements; diamonds represent decision points. Programming Language: How you would finally implement an algorithm on a computer. Block diagrams: Yet another way. Move from natural language to pseudo-code to programming language.