.
Profile URL | https://www.quora.com/profile/Bill-Paseman |
Question | Answer | Date |
---|---|---|
A friend got an idea to found a startup and asked my help. she has some money to invest in but not much time to dedicate (she has a full work), I have time and the knowledge to get this idea real. How should roles be decided upon? | She has two roles: angel and consultant. You represent the company in both negotiations. The way to establish a fair price in both angel and consultant roles is to have more than one bidder. If you cannot get angel or consultant interest beyond hers and decide to "go ahead anyway", the chance of success is a lot lower since neither the angel amount or consulting rate has been vetted. The amount of equity the angel gets is determined by whatever term sheet you negotiate. The term sheet will determine the common and preferred price. The amount of common stock the consultant gets is determined by the consulting agreement. Often the stock a consultant gets is determined by their rate/preferredPrice. Both negotiations are critical, but it is especially critical to determine how much you will give out for the second part. If the rate is "per hour" and the consultant goes over, then expectations on your part will not be met. You will "walk", she loses her money and everything will fall apart. You ought map out the expected tasks (this must be done very well) and have her either bid per task; or bid a "not to exceed" per task so that there is a cap on equity she will earn. As a side note, in my opinion, the value she gets for "the idea" is being able to bid in both roles. If that is not enough, the deal is way too screwy and I would walk. | 9/26/2015 |
After Andrew Ng's ML course should I do Geoffrey Hinton's neural network course before doing deep learning? How much neural networks knowledge is good enough to start doing deep learning? | You can probably start applying it after Andrew Ng’s course. You have a better shot at understanding it if you take Hinton’s course. | 2/14/2017 |
Am I able to study medicine without doing math at high school? | So long as you don’t dispense medication (requires measurement) or look at clinical trials for very sick patients (requires understanding statistics) or do research in applying machines to data interpretation (helps to know calculus) or want to talk to all your other physician colleagues (who did well in math), then you should be OK. | 1/1/2018 |
Any good practical tutorial and examples regarding how to train and tune the recurrent neural networks? | 3/23/2016 | |
Any ideas for an AI project? | Answer 2: There is an MIT AI Memo or TR from about 1977 that discusses machine generated disney-style story generation. The “princess” usually appears on stage with grand sweeping motions and the villain also has his own gestural identity. If the robot can recognize gestures, it could act to fill in missing roles. E.g. If the robot recognize the gesture of the person in the room as being a villain, it could become a princess and sing “some day my prince will come..” and move back and forth in sweeping gestures. If, on the other hand, it recognizes a princess coming on stage, it could adopt the role of villain. If you create something that is solely a princess, it could wear a dress and spin faster or slower based on the speed. Be sure to use a high pitched cartoon voice. The effect will be better. Simpler to implement, have an LCD “face” that brings up different disney princesses depending on the aria sung by the villian. E.g. if it recognizes Gaston, the face of “Beauty” pops up. I say this assuming that spectral analysis show easy to identify differences between the different villains. Step zero in this my be to have the villain announce themselves. “Beauty, it is I, Gaston!”. Worse case, you record the audio from the film and use that as the trigger. The nice thing about this is that there are so many villain/heroine pairs out there that you can try all of them and see which two your recognition algorithm finds farthest apart. So.
You have now created a mechanical actress that can play any disney role, will not forget its lines or step on the lines of other actors. ======================================================= Answer1: If the drone is airborne, you can probably accomplish most of this by following the script given in aero_astro 16.30-16.31. 16.30 is s semester course, so make sure you have enough in place that you don’t have to recreate everything from scratch. Is the input verbal? If so, does Siri have an API? If so, put a fair amount of thought in error handling. E.g. if your professor tells it to “go to hell”, you could say, “Thanks, but I’m already enrolled in your course”. Perhaps the easiest thing to do, given that the semester is started, is to break your project into demonstrable steps:
| 2/28/2017 |
Are Angel Investors put off or motivated by young, first time entrepreneurs that pitch their ide? What advice would you give them? | Neither. I am motivated by someone who takes the project seriously and has shown real progress against these milestones P&A: Pitch Checklist regardless of the age. | 7/4/2016 |
Are humans just a bag of chemicals and are all our thoughts, emotions and actions just a bunch of chemical reactions? | 2/25/2017 | |
Are terms agreed upon between founders via Email binding upon founders if disputes arise before signing an operating agreement? | If the terms are cast into a Contract, and the contract is valid (signed by both parties among other things, note that state laws apply), then yes. Generally you need a lawyer to make sure the contract is valid. | 7/21/2016 |
Are there any books that take calculus from the philosophical side? | Though not calculus specific, I enjoyed The Mathematical Experience: Phillip J. Davis, Reuben Hersh: 0046442929684: Amazon.com: Books | 11/5/2017 |
Are there any good science fair project ideas based around murder and/or serial killers? | I can think of lots. Mostly based around statistical correlation. E.g. murder rates through time, by geography, before/after particular pieces of legislation were passed (e.g. did capital punishment increase/decrease incidence), by country (e.g. in Mexico through time, lots of people were killed as the state became more corrupt. Many of the serial killers in this case were cartel members). Despite the media coverage, I have read that the world is now safer than ever before. Is this true? What statistics were used to back up these claims? Have murder rates gone down in states that legalized marijuana (stoners don’t kill) or up (Stoners move to crack and steal/kill)? I once (accidentally) hit a guy on a motorcycle in Amsterdam. Turns out he was a homicide cop. They have a drug park there. I asked him his position on marijuana legalization. He was against it because he said it made his job harder. Why? Ask some homicide detectives to find out cause/effect and check it out with statistics. | 1/10/2018 |
Are there any well-known companies or start-ups that began in a bar? | Compaq and Southwest are the apocryphal examples (Napkins: Where Ethernet, Compaq and Facebook's cool data center got their starts) I also heard that Pet Rocks were actually conceived of in a bar. | 1/5/2015 |
Are there crowdsource businesses where people work in teams on projects? | Coursera | 3/29/2016 |
Are there examples of life sciences orgs achieving results in drug discovery via machine learning that wouldn't be possible without machine learning? | Judge Schonfeld (Real-Time Evidence Based Medicine) announced in November 2015 that his group had discovered that a cancer drug could also be used to treat Ebola. This leads to the world of Drug discovery by Drug repositioning - Wikipedia. The basic idea is that there are lots of drugs out there that have passed phase 3 trials (I.e they don’t kill you outright, dosing is known and they have some non-fatal effect on the body) and all these drugs have side effects (The most famous example is Sildenafil - Wikipedia, which originally was used to treat pulmonary arterial hypertension, but was then discovered to treat erectile disfunction and was rebranded as viagra.) So you put all these drugs in a database and run algorithms to see if they can be used to treat conditions other than the primary target. This is an area of active interest since 25% of most drug company revenues come from repurposed products (Therapeutic Drug Repurposing, Repositioning and Rescue Part III market exclusivity using Intellectual Property and regulatory pathways.) There are lots of ways to do this at many different levels (e.g. [1502.02072] Massively Multitask Networks for Drug Discovery). However (I suspect) most of this work is being done internal to drug companies since they are most interested in leveraging their own drug portfolio, not curing a disease using someone else’s drug. | 2/20/2017 |
Are we destined to be conformists as we miss the opportunity train that properly nurtures and develops the next big idea? | A few thoughts.
So. What to do? Best thing that works for me is to listen. a lot. to aliens. | 11/11/2017 |
Are you treated differently when you wear a Stanford (University) sweatshirt in public? | I live in Silicon Valley and sometimes wear a Stanford sweatshirt. Every now and then I'm asked "Did you go to Stanford?" My (truthful) response is "No, actually I went to MIT, but I don't like to brag". | 10/8/2015 |
As a developer should you be writing code on a Macintosh? | Ability to write os level scripts becomes handy at some point. Unix variants have good scripting capability. The Mac runs a Unix variant (MACH). Ubuntu is another. | 12/15/2017 |
As a doctor who programmes and is involved in technology research, how can I start learning about machine learning/deep learning? | See Bill Paseman's answer to What are the most important deep learning algorithms? In which order should I learn them? | 4/23/2015 |
As a man, how much are you willing to pay to have a barber come to your home and give you a cut or shave? | I have a woman do it. She also washes my hair in a basin (which I do not have at my home). I actually enjoy going out and have lunch afterwards usually. | 4/25/2016 |
As a startup founder, seems like everyone wants equity as an option of payment- who should actually be getting an option to own part of the company? | http://www.barnesandnoble.com/w/... Lists some equity division strategies based om deliverables. | 10/14/2015 |
As experienced software engineer how to get involved with machine learning or neural network development? | 2/20/2017 | |
As for machine learning or deep learning, is there the phenomenon that the model works well but the designer cannot explain the reason? | Neural networks are black boxes, They are an almost total surrender of mechanistic reasoning to statistics. The explanations then of why they work (or don’t) are based on statistics, not aristotelian logic. From the video below: “the sample set is too small.” “The bias is too big.” “The variance is too big.” Not “The system does not understand green chihuahuas in the rain”. As such, I suspect your question is non sequitur, the designer -never- understood the reason why the model worked well (or not) in the first place. At least, not in non-statistical terms. The network develops its own way of answering the question (Note in particular Dr. Ng’s “Phonemes are a fantasy of Linguists” comment. ) A wonderful example is from Detecting Tanks, where Ed Fredkin saw a presentation describing a network trained to detect tanks, and remarked that the network actually was distinguishing cloudy and non=cloudy days. This video is listed here: Bill Paseman's answer to What are the most important deep learning algorithms? In which order should I learn them? | 3/9/2017 |
At what point does a startup start looking for a CMO? | I was told to hire a marketing guy as a cofounder. Employee number 3. | 11/22/2015 |
At what point should artificial intelligence be afforded the same rights as humans? | When it starts paying taxes | 2/16/2017 |
Can anyone give me any BIG ideas that I could do for my personal project (a year long project)? If you do I would be truly thankful | Any particular field of study? How about a cure for cancer? You could get dogs from the local pound which are dying of cancer, feed them a ketogenic diet and see how their tumors progress. | 5/19/2017 |
Can genetic programming, metaprogramming and deep learning be incorporated into the same programme? | 12/21/2017 | |
Can I change my science fair project in between levels of competition? | This document http://www.mcsta.org/wp-content/..., lists some phone numbers you can contact to ask. | 3/11/2017 |
Can I get the relationship hierarchy of a product using Ontology? | Check out protégé | 3/15/2016 |
Can I use complex numbers for the weights in the deep neural network? | Could you give an example? What activation function you would use? (This is the sigmoid, tanh, relu part of the network that determines if the result of the prior layer is passed on to the next layer.) So, Sangnam Park gave an example activation function which, (I believe) converts complex to reals. So if that is the case, let’s pick a pretend application; say signal processing. Suppose we take an FFT of an incoming signal. One approach would be to simply separate the Amplitude and Phase of the FFT output and feed them in separately and let the network adjust the weights. While that is more flexible, if you can come up with a single weight for both IM and RE, you could cut the number of inputs in half and maybe reduce the amount of training time. What would be fun is to split the RE/IM parts and, while not definitive, see how close the corresponding weights become after training. | 12/16/2016 |
Can I use deep learning to analyse text data? What are some good resources to learn deep learning from scratch? (Resources for Python will be helpful) | 1/17/2017 | |
Can the NVIDIA Jetson detect bus passengers' wrong movements in real time, from the CCTV footage? | I suppose it depends on the resolution and frame rate of whatever you want to detect. Take some footage at jetsons resolution and frame rate and see if a human expert can detect ‘wrong movements’. If so, you have your answer. | 9/14/2017 |
Can we as humans rationally distinguish between science and religion? | If you view the scientific method as "opinion", then no. However, 300 years of steady scientific progress would seem to indicate that there is a difference between the two. | 11/19/2014 |
Can we control metal by strong magnetic fields using AI to produce it whenever necessary? | Well, not my area of expertise, but I believe theories of magnetism involve the presence of d orbitals, and not all metals possess these. Not sure how ai would figure in. | 7/27/2016 |
Can we ever make an artificial replica of the human brain using algorithms of machine learning, mainly neural networks? | Check out What are the key differences between artificial neural nets and real neurons of the human brain? | 11/17/2014 |
Can you give me some information about recycling ideas, machines and succeeded projects.? | After inventing the light bulb, Edison created an iron ore extractor based on magnetism. I believe many automated waste separators extract iron the same way. In addition, if you grind up the waste finely enough, you can use density to separate the plastic from other materials. FInally, if the waste is burned hot enough, it can be used in energy generation. | 10/16/2016 |
Computer science/Aviation final year project ideas? | I think it is a reasonable question to ask. There are lots of aspects. E.g. I went on a Boeing tour several months ago, and they showed their assembly line of planes. The newer ones are smaller, made of composite materials (vs. aluminum) and have computer controlled devices everywhere (e.g. the passenger window glass). Hobby Drones are steadily improving. Go to Fry’s and look at the latest model features and compare to some of the older models. This will give you an idea of feature trends. Tesla develops drone components for its cars (electric motors, navigation and batteries). Look at projections for improvement from the analyst reports and assume the components can be repurposed for drones or light aircraft. This will give you an idea of component trends. | 10/17/2016 |
Computer science: If you want to be an IR or an AI expert, how skilled in hardware should you be? | If you want to be skilled, you probably can get by without it. If you want to be an expert, you will need to under stand IT, how to parallelize your work using GPUS, etc. Look at Andre Ng’s initial comments ( ). | 12/20/2016 |
Computer science: In what ways might artificial intelligence help an entrepreneur? | I was a grad student in Steve Ward's group at MIT from 1978-1980. My Masters thesis explored the application of AI to music composition. Although I framed my work in terms of generative grammars, a colleague, David Levitt noted that jazz improvisation could be structured using Waltz-like Constraint networks. I spent a few weeks adapting this to species counterpoint. Fast Forward to 1992. I had a contract to create a "configurator" for WANS, and later PCs. I noted that I could adapt the same constraint code to configure these products (as well as insurance policies, financial plans, fork lifts, etc.). I formed a company called Calico which commercialized this. It went public in 1999. I then retired. Two Notes: 1) I actually have degrees in ChE (Petroleum Engineering) EE (Signal Processing) and EECS (AI). And after all of that, applying AI to music composition is what made my fortune. 2) Probably having a broad background helped me be a better entrepreneur than knowing AI. My attitude was that I could pretty much do any type of engineering. Engineering is the commodity. What is in short supply is good, expensive, solvable customer problems. A broad background allowed me to ask the important question: "What is your most expensive engineering problem and how much are you willing to pay to solve it?" | 2/19/2016 |
Considering A.I. could write code (programs) in any language or tool if it’s generally intelligent, would it even care about readability of programs? | Compilers don’t. | 2/14/2017 |
Could AI become advanced enough to be indistinguishable from humans? | Its been done. Long ago. When PARRY Met ELIZA: A Ridiculous Chatbot Conversation From 1972 Joe Weizenbaum wrote the original ELIZA. I actually had his office after he left. He once read a gushing review of “Parry” (above) that said that PARRY provided a wonderful opportunity to study the mechanisms of paranoid schizophrenia . Dr. Weizenbaum wrote a biting response. He said that he had just created a program to simulate a far worse malady: autism. It consisted of one line: While (True); input X He mentioned that it was so good that it would function even if the computer wasn’t turned on. This anecdote does a good job of pointing out the difference between simulating the I/O characteristics of something and creating a higher fidelity model. Critical to note though that even creating a high fidelity model is not the same a creating a ‘being’. | 7/7/2017 |
Could an error in an AI's code make it self aware OR allow it to evolve into something else? | I was an MIT grad student in 1979 and was listening to Marvin Minsky field questions in one of our AI classes. One student asked “What would you do if you arrived one morning and all of a sudden, your computer was intelligent”. Unbidden, I answered “Unfortunately, we never seem to have that kind of luck around here”. Minsky laughed, said “of course that is the REAL answer”. | 1/6/2017 |
Could one start a small business having knowledge of Natural Language processing? | Well, how deep a knowledge? Let me tell you a problem I have, and if you can solve it, let’s discuss a price. Home - ClinicalTrials.gov has a database of semistructured data. I need to narrow down the descriptions of particular conditions to a small selectable subset. Currently, I have to hand code a recognizer for every single goddamned disease set. And my work is not that good. E.g. if the text says “you can participate in the trial unless you have condition ‘X’”, my code happily says that people with condition ‘X’ are eligible for the trial. So. If you can do stuff like that, then I think you have the basis for a good consulting practice. | 9/29/2016 |
Could you suggest me some best website for Digital Logic projects and innovative ideas? | 40 years ago I looked at books by don Lancaster. E.g. Ttl cookbook, and used those to imagine stuff like digital music synthesizers. Nowadays maybe make magazine and various maker Faries. | 9/26/2016 |
Do any rich people have associate's degrees? | Well, Steve Jobs and Bill Gates were dropouts and had no degrees at all. Do they count or are you interested in Associates degrees in particular? | 10/10/2015 |
Do colleges offer machine learning course to undergraduate students? | Look at the coursera course list. It will tell you some colleges that offer these courses (e.g. Stanford) | 12/14/2016 |
Do I need a supercomputer to run artificial intelligence (human level) on it? | Re PC, Yes, if you believe the Church–Turing thesis - Wikipedia (which has been around since the 1930’s) and speed is not an issue. To Emmett’s point, a -natural- model of human intelligence has not yet been puzzled out. Speculations here Bill Paseman's answer to Life originated from chemicals, Abiogenesis, and evolution did the rest of the work. All of our thoughts and emotions are simply electrical charges and chemical reactions in the brain. Are we just bags of chemicals, moving around, and nothing more? | 1/15/2017 |
Do I use "variables and controls" in a mathematical science fair project on statistics? | You could. What is the project? | 12/29/2016 |
Do investors inquire about entrepreneurs health? | For each initial (multi-million) investment I got for my two companies, the funders (Kleiner perkins and Mayfield) sent a doctor to my office to do a physical which included a blood and urine test. | 6/30/2016 |
Do we have statistics on the number of academic papers published in AI in the past two years by top Silicon Valley companies namely, Google, Facebook, Amazon, Microsoft, IBM, Adobe, Baidu, Tencent, Apple, Yahoo, NVIDIA, Salesforce and Quora? | You can probably get an idea of the ratios by doing queries on arXiv.org e-Print archive E.g. https://arxiv.org/search/advance... gives 14 results. Google scholar gives 7,700 references which can be culled with keywords | 5/4/2018 |
Do you have any project idea that can improve the mutually shared places in the cities that have been misconceived? | 9/26/2016 | |
Do you usually think it is easier to complete a project individually in a group? | I assume the question is “Do you usually think it is easier to complete a project individually or in a group?” It depends on the individual and it depends on the group. An obvious example. Suppose your “project” is to lose weight, get off drugs, become more fit, etc. Research indicates that if you are in a group of like minded people, it is easier to accomplish a goal. E.g. better to be in a group of people trying to go straight than people who love doing drugs. Consider programming. In my experience, it helps to have a domain expert available, otherwise you (the naive programmer) have to learn a new domain and the pitfalls therein. Slow going. But both of you have to be open to one-another. | 6/18/2017 |
Does an idea have to be in steps? | An elaboration of Daniel's point. An idea does not have to be in steps. But a BIG idea is most likely implemented in steps. There is an old quote. "Don't mistake clarity of vision for shortness of distance". You may be able to clearly see a mountain on the horizon and decide that you want to walk to it; but if it is hundreds of miles away, miles high and there are several deserts between you and it, you most likely have to have a plan beyond just "walking" to reach it. Since things like ideas, plans and steps have a tendency to be confused, and this confusion results in lost time and effort (especially among groups), Planning methodologies often formally distinguish them. For example, from GOSPAPlanning.com UK, US, Europe
| 3/19/2016 |
Does genetic programming belong to deep learning? | You can use genetic programming to control deep learning parameter selection. E.g. rhiever/tpot | 12/21/2017 |
Does the city of Bielefeld actually exist? | No. It is a conspiracy of linen weavers. | 1/27/2017 |
Entrepreneurs: What do you think of a course/ebook that teaches you how to negotiate? All secrets and tips to convince anyone and having a deal. | I think its a great idea. If you were able to organize the course around some social network, you could even assign homework where the students do role playing with one another, trying to convince them of various arguments. | 5/4/2016 |
Every time I begin a personal project with an idea, I fail at continuing it. What can I do so I won't change ideas fast? | If you have a degree of some kind, you probably had to focus to complete it. Whatever methods you used then, try applying them now. If not, create a plan for each project and set small milestones. Try to arrange the milestones so that the key feature you want to demo is available after a short period of time. Continually demo to an external party and see if you still feel good about the project after finishing the demo milestone. You may find these three examples helpful. Windows: In the early 80’s at Daisy systems, the president asked me to create a windows system. The OS group had several people working for a year and hadn’t completed anything. Much of the work in windowing system was handling updates to overlapping windows. So the first thing I did was force any window that was updating to the front so I did not have to deal with the problem. I also had my window system only work inside of one application (a discrete event simulator) so I didn’t have to deal with multiple processes. I leveraged existing packages (a bit level “screen region” mapper and a symbol table package). I had the first version working in 3 days. The president liked the crude demo so I spent more time on it. Adding borders, providing control key commands to grow/shrink/create/delete windows and associate signals with them, etc. That took 3 weeks. The next demo went well and I turned it over to another guy to add a test suite , fix bugs, etc. It took him 3 months. This window system-inside-a- simulator worked well enough that the sales team could sell it. This gave the os group time to finish their version. Simulator: I had gone through this 3 day, 3 week, 3 month cycle before while managing the creation of the discrete event simulator. The simulator had three main parts: a console, a simulation engine and a compiler that turned the schematics into something that the engine could simulate. I first had to create a command console, so I reused the schematic editor console that I had spent several months on and simply switched out the schematic editor’s command parser. That took one day. The one-day demo simply consisted of entering a sequence commands that we would eventually have. The commands did nothing. I demoed it to the VP of Engineering and he was very happy since although it did nothing, it did it after just one day. I then hard coded a simple 4 nand-gate xor circuit for the engine to simulate and added debug commands so the engine developer could dump the network and queues. That took three days. (Another lesson here is that you will need a specialized debugger/structure dump for any complex product implementation. The earlier you build it, the more you can use it. So build it first and continually add to it.) Hard coding the xor circuit into the simulator decoupled console/engine development from the compiler development and let me demo the console and let the engine developer debug his engine before the network compiler was even coded. The next demo simulated a few steps of the 4 nand gates and did a hex dump of the state array to the screen after each step. No schematic compiler at that point. The first compiler took 3 weeks. Three months to work properly (During that time, the engine added additional behaviors beyond ‘nand gates’ and the console moved from a hex dump to displaying waveforms.) Spreadsheet: I once had a consulting assignment where I came into a spreadsheet company which was in complete disarray and had had a series of VPs of engineering, all loathed by the staff. My mandate was to get things on track. They wanted me to be the next VP. I said “no”. I asked for a VP’s consulting rate, but asked to be assigned to the “in memory database”. Since the DB had no user level visibility, I politely asked each of the other groups: UI, reasoning engine, etc. to please provide me with an interface so that I could demo my db to the president. The first demo was simply to bring the product up, show a prompt and quit. (They hadn’t even gotten that far yet.) But this demoed a complete product integration. Next we showed a few cells that came from the database. And so on. My biggest contribution here was to identify the two key -emotional- issues: 1) The developers hated bosses, so I didn’t become one and 2) ultimately, they wanted to show off their work and demos provided an opportunity to do that -in the context of a whole product-. Priority Algorithm: Now, here is the priority algorithm. Create these 3/3/3 schedules for all your projects. The ones you can’t do this with, Defer. Demo the 3 day milestone for each project and see which one you (or your customer) gets excited about. Again, the project’s one killer feature needs to be there after 3 days. If you can’t conceptualize, implement, demonstrate it; find a project you can. The implementation will have a LOT of stubs. resist the temptation to fill them out for the 3 day milestone. Again, the idea is to demo the one ‘killer’ feature. Philosophical BS: Jim Barkesdale (CEO of Netscape) used to say “Never confuse ‘clearness of vision’ with ‘shortness of distance’”. When traversing that distance, recognize that the key issues here are emotional, not rational and that people (including yourself) can maintain faith and hope for only so long. So focus on short, clear milestones to justify faith and increase hope. Past the three month milestone, you still need to set proper milestones. You might find this helpful: How do I stay motivated and on track as a 'lone wolf entrepreneur'? Just saw this quote from Napoleon Bonaparte: “A leader is a dealer in hope.” In my experience, this is true even if you are just leading yourself. | 5/22/2018 |
For a European startup, what is the best way and moment to relocate to Silicon Valley? | Well, what do you hope to find here? E.g. Venture funding is very sensitive to the Nasdaq, so if it is funding, moving when the Nasdaq is rising, Venture capital is pumping up rents and unemployment is low (e.g. about 8 months ago). If it is getting cheap help and cheaper space, it is about 3-6 months after the Nasdaq plunges, funding dries up, vacancies increase and people are willing to work for less. | 10/16/2014 |
Harvard University: What are the best research-based AI classes at Harvard for someone looking to get a PhD in Computer Science after graduating? | I believe that Harvard students can attend MIT classes for credit. (Eg Guy Steele) You might look into graduate seminars there. In addition, it will allow you to scope out some potential advisors outside of Harvard. | 10/25/2014 |
Has Apple's ResearchKit been adopted for clinical trials research or patient recruiting? Can anyone share any case studies? | Apple's ResearchKit generates reliable health data — at least for asthma patients The Asthma Mobile Health Study, a large-scale clinical observational study using ResearchKit | 10/24/2017 |
Have you started a successful startup whilst employed? | Employed as what? If you are getting W-2's (in Silicon Valley anyway), there is a good chance that you have something in your employment agreement that discusses what "work products" (e.g. inventions) belong to you and which belong to your employer. In my case, it was pretty strict. However, nothing in my employment agreement forbade me from quitting or from looking for (non-competing) consulting work while I was employed. So I identified and closed my first customer, quit my job and starting a consultancy/contracting firm. This solved the following problems: 1. It brought in 1099 income (meaning I wasn't employed by my client, or bound by their employee agreements) 2. It allowed me to negotiate a a contracting agreement with my clients with a less restrictive NDA. 3. It allowed me to identify my market by working in close contact with my future customers. This was important since I had started another company earlier that failed due to inadequate marketing. 4. The (future) clients were only interested in hiring me if I had good references for my (then) current employer, which I had. 5. This approach kept me from terminally pissing off my (then) current employer, since I didn't break any contracts I had signed. As you mention, time management is important. I was told by a friend who started a consultancy that fully 30% of his time was spent doing lead gen (phone calls, blogs in your area of expertise, etc.). If you have 50% utilization, you need to shoot for a consulting rate that annualizes to >= 2x your current salary (e.g. $100K fully burdened salary implies at least 2000 hrs/year x 50% x $100/hr). That leaves 20% for product development. One way to get more is at step 2. If you can convince the client that as a consultant, you are bringing in your own work product and can't assign it to the client, there's a good chance that a lot of the -utilities and tools- you create can be re-used for the next client. Of course, you will not be able to keep anything that is proprietary. (Side note: During this phase, the bad news for me was "maintaining focus and keeping up the pace" was not always easy. The good news for me was that I got to test my theory that I could do better on my own that working for someone.) In my case, after 2 years of market identification and 2 years of development, I had enough surplus receivables that I could bill for them all on January 1st and have my year's "nut" covered. This allowed me to switch from selling consulting to selling product. But that is another story... | 11/6/2014 |
How actions are taken in a q-learning problem where the q-function is a neural network? | I suspect it depends on the domain. However this approach was used to automatically create an Atari video game player (http://www.nature.com/nature/jou...) and so actions in this case would be cursor control and “fire” buttons. | 1/13/2017 |
How are artificial intelligence and neural networks related to petroleum engineering? | You could use them to analyze seismic logs, well logs, satellite photos. Any non-structured visual or auditory information. | 6/12/2017 |
How are bounding boxes in images formed in computer vision? | There are different approaches. Do you know what you are searching for? (E.g. "Robust Face Recognition via Sparse Representation" http://research.microsoft.com/pu...) The Cutting Edge of Computer Vision assignment at Stanford from a while ago shows another approach. Richard Socher's video at BAMMF shows a Deep Learning solution for image search united with language understanding. Although he does do image identification, I don't believe he uses bounding boxes at all. | 3/28/2015 |
How can a carbon capture and storage model be made for a high school science fair project? What are some tips? | You might try coming up with several demonstrable models and put a placard in front of each to show the resource flow and the math. Here are a few. Amazon.com: SodaStream Jet Sparkling Water Maker Starter Kit, Black and Silver: Kitchen & Dining - How much CO2 can you actually pump in per unit volume? What happens when the container is breached? How long can you keep the container under pressure before “natural forces” cause it to leak away its contents? How much CO2 out there must be sequestered in order to change climate dynamics? How much CO2 can all the suitable aquifers are there actually accommodate? How much energy is required to do this? If the energy is created by burning carbon, how much coal/oil/natGas efficiency is lost by CO2 sequestration respectively? Creatures that create shells vs. making concrete - How hard is it to sequester CO2 as calcium carbonate? Much of that comes from limestone, which comes from sea creatures. This form of sequestration is very stable, but takes longer than the other approaches. In fact, a lot of CO2 comes from decomposing limestone into cement using heat. (Cement is used in making concrete, which also generates CO2). Suppose you simply made less concrete, leaving existing limestone alone. From a chemical balance perspective, you could view that limestone as being “pre-sequestered”. So how much “pre-sequestration” would you get if you simply replaced concrete with another building material? E.g. asphalt roads (which also generates CO2). Grass vs. trees - I read somewhere that prairie grass is more effective at capturing carbon than a tree is. However, both eventually die and, at some point, release CO2 back to the atmosphere. This is happening to the tundra in Alaska, which is adding to the CO2 load now that the snow cover is gone. At what rates can you sequester CO2 for each approach? Both need water. How much energy do you need to get them water, and how much carbon dioxide is generated by creating that energy? In doing the balance equations, for non-carbon based energy generation approaches, how much energy is used to create a solar panel (be sure to factor in a panel’s usable life) or the hoover dam (made of concrete- see above) or a pound of U232 by the TVA? I once heard that if you looked at the energy balance eqns used to separate U232 using hydro-generated power by the TVA, U232 was not so much an energy source as a battery. I.e. it took 1 kw of hydro energy to generate enough U232 to generate 1kw of power. | 2/13/2018 |
How can a person develop his mind to solve real life problems? | Ayn Rand has a pretty good essay on this. see Bill Paseman's answer to What is the task of philosophy? As I note there, all philosophy boils down to one question “What do I do next”? Since you mention ‘real life’ problems, I’d first suggest that you don’t overthink and continually engage the real world directly, but consciously. Most people engage the world unconsciously. By that I don’t mean that they don’t know what to do next, but rather they DO know and are unaware of any alternatives. So for every action, think through alternatives. Second, I tend to think in terms of Control theory - Wikipedia. This allows me to step outside of myself, away from my emotions and so deal with my approach to the world more objectively. Here is some of what control theory has told me.
Hope this helps | 5/14/2017 |
How can AI be used in conjunction with blockchain technology and what are the potential implications? | You could potentially use it to track illegal bitcoing commerce. Eventually discovering who bought what for how much. | 8/3/2016 |
How can an entrepreneur find seed money? | Become a consultant. You will only get people who have problems bad enough that they will pay for the solution. This helps you determine the market, product and also give you initial capital to boot. | 3/31/2018 |
How can artificial intelligence help winning elections? | Bill told Hilarie to pay attention to Wisconsin. She and her advisors laughed him out of the room. They felt that he was concentrating on his voters, not hers. So here is an ai. Write it on a piece of paper and hand it to the next democratic presidential candidate. “Shut the hell up and actually listen to what bill Clinton has to say”. Perfect piece of code. No computer required. | 1/28/2017 |
How can classical AI techniques involving knowledge representation be taken further? | How indeed. You could help Cyc or Watson pull a “Hinton” and outperform some existing benchmark. (e.g. 2011 Jeopardy). One thing about this though, is that it all seems to be closed source. I asked 5 different IBM sales reps about their work with a Boston hospital and was unable to get ANY info. Or. You could support the other team. Synthetic data synthesis (see video below), though Dr. Ng may say “phonemes are a fantasy of linguists”, his still needs -labeled data- and those labels pretty much stem from the ancient greek concepts of logic and knowledge representation taught in the 70’s. You might want to look at Richard Socher’s Metamind work (before he joined Salesforce) as well. I’ll add more if anything occurs to me. | 3/10/2017 |
How can I appeal to the judge's decision in the Google Science Fair? | I don’t know. My kids participated in several science fairs and I know that some projects were trendy. As I recall, the winner had an idea of how to use orange peels for water retention, yes? I can see how that would appeal to judges in drought stricken California. However there are many examples of unappreciated ideas which worked out well (e.g. FedEx). If you believe in it, you might show it to others and see if it can be explained more clearly. Then extend it and resubmit it to another science fair. If it is really good, turn it into a company. Who knows. Maybe it will be acquired by Google. | 9/30/2016 |
How can I be sure people won't steal my idea during a pitch? | Submit a provisional patent beforehand. | 7/30/2017 |
How can I become a successful entrepreneur at age 15, and make adults take me seriously? | Young Michael Dell (Dell Computers founder, author of Direct from Dell: Strategies that Revolutionized an Industry (Collins Business Essentials): Michael Dell, Catherine Fredman: 9780060845728: Amazon.com: Books) aggregated paper routes by opening new territories, recruiting subscribers, staffing the routes and taking a monthly cut when collecting fees. He got in a conversation with a school teacher who saw him drive a BMW to school. Turns out Michael made more per year than the teacher did. The teacher was upset, but “took him seriously”. Net, do something entrepreneurial that makes you more money than them. | 10/30/2017 |
How can I catch venture capitalists? | Give them a no-fail pitch
| 12/10/2017 |
How can I clearly and deeply understand the mathematics beyond the different neural networks models? | What are the most important deep learning algorithms? In which order should I learn them? | 3/4/2017 |
How can I create successful project if idea isn't unique? | You can execute more cheaply, faster, with better quality. | 5/17/2016 |
How can I eliminate $50000+ debt and same time start my own side business at 43? | There are several organizations which help pay for surgery if you can’t do it yourself. Forums like SmartPatients discuss questions like these. Get your sister to reimburse her surgery through these channels and pay you back. Keep meticulous track of the costs and payment sources and write a book about it, offering advice to others on how to get back on their feet. | 6/20/2016 |
How can I find problems and capitalize on the solution? | Become a consultant. You will only get people who have problems bad enough that they will pay for the solution. This helps you determine the market, product and also give you initial capital to boot. | 5/20/2017 |
How can I get a computer science internship as a high school senior? | You can do that. However the two red flags to me in a role as a hiring manager are: 1) Why aren't you interning with the mentored companies and 2) The fact that you did not finish your "useless programs". so I'd suggest a) Go back to the mentored companies and say "Having worked with you in the past, I couldn't help but observe you have problem X. I'd like very much to help you solve X in the following way." and then give details. X may no longer be a problem; in that case ask "what is a problem?" and help them solve that. b) If a few projects are not far from completion, finish them. In particular prioritize ones that may be of interest to companies you want to intern with. This second point reminds me of my first lesson in graduate school at MIT. A meeting was called "early on" requesting attendance of all the grad school faculty and new students. In the front were arguably some of the best young CS minds in the world. In the back were many of the men who defined the field: Joel Moses (Symbolic Math), Marvin Minsky (AI), Barbara Liskov (Languages), Computational complexity; the whole shebang. I was a bit puzzled. What topic could possibly cut across the an audience that varied in skill, interest and experience? What the speaker said has stuck with me. "To the new students. Congratulations. I am here to emphasize a mundane point. If you want a degree, you need to write a thesis. To write a thesis you need a research project. The project needs to be special to you. You need to focus on it from its beginning to its end. It needs to maintain your focus when it is popular and when it is not. Because without finishing the project you will not finish your thesis. And without finishing your thesis, you will not graduate." So. That was the common point of interest to an audience of the best of the best at MIT. To young and old, experienced and not, pioneers and followers. Finish it. Not because it was fun but because you committed to it. Now, the subtext to me was that even MIT guys (grad students and their advisors) needed to be reminded of this occasionally. Two smaller anecdotes. I started my original 4 year degree in chemical engineering. I developed an interest in Electrical Engineering and so finished courses in both (in 4 years). I did not get all A's. But finishing helped me get into MIT grad school. In my first Silicon valley job, I worked for some Israelis who complemented me on my MIT degree. I said "You realize that I didn't exactly get all A's". They said "Yes, but you did get through". That's what they were looking for. That's what I gave them. ANd that is what I continued to deliver in a business context. Results. If money is your motivator. That's what helped me make my first million at my first company (back when a million meant something) and that's what has helped me make subsequent millions. I envy you. You are young and it sounds like you have a lot of what it takes to do well. Good luck on your journey. | 11/11/2014 |
How can I get along with math? | This might help: The Mathematical Experience: Phillip J. Davis, Reuben Hersh: 0046442929684: Amazon.com: Books | 1/24/2018 |
How can I get funding to finance my project? | If it is it a hobby, then you’d probably be funding it yourself or via crowd funding. If you want to sell a product or service, I did this by starting a consulting practice to gather requirements, create a prototype and get initial funding (from the consulting fees). Once I got traction, I got VC funding. Teknekron was expert at this. Checkout the Harvey Wagner references here: P&A: Pitch Checklist | 3/6/2018 |
How can I prove that I donated money to charity, without revealing personal information? | Pay cash. The charity should give you a receipt.... | 10/10/2015 |
How can I share my side project (forum, etc.)? | Build software better, together OK, From the comment, the question is not about sharing, but about promotion. Not my area of expertise, but traditionally, you’d
| 6/20/2017 |
How can I start applying deep learning in robotics? | Probably in object recognition. Some resources: What are the most important deep learning algorithms? In which order should I learn them? | 7/27/2016 |
How can I start programming machine learning and artificial intelligence? | The best resource listed there for introductory machine learning is Andrew Ng’s coursera course. | 2/20/2017 |
How can I study for the finals of the first university year? | The other advice given here is solid. One other point, have you passed the ‘drop’ date yet? If you you were lax in the beginning of the semester, have learned your lesson and are facing an insurmountable wall, then consider fessing up and talking to a counselor about dropping a class. You may need to take another semester, or make it up during the summer, but that is probably better than going through the next three years with a bad understanding of the basics. | 3/12/2018 |
How can I study speech recognition with machine learning? | Richard Socher - Home Page from (MetaMind - Deep Learning for Enterprises) outlined his language understanding approach at the fourth talk at the November 20 2014 BAMMF forum at Parc. Here is the video (Bay Area Multimedia Forum - 20 November 2014 - Richard Socher). His approach is interesting in that it:
| 1/17/2015 |
How can I take my Arduino project to the next level? | I created a medical device (a variation of a pulse oximeter to detect hemoglobin) using a Due. Choosing Due was important since it has two “real” D/A converters. You could also figure a way to use an Uno to implement a deep learning “recognizer”. (It would have to be pretty small though - Bill Paseman's answer to How do you implement Artificial Intelligence in an Arduino Project?). Since I did the above analysis, various articles on this have come out: Arduino Neural Network Good Luck! | 12/4/2017 |
How can I understand the elementary constructs and assumptions of mathematics? | A broad brush approach (- proofs) is in The Mathematical Experience A small book, poorly reviewed, but with a historical POV is The Great Mathematicians: Herbert W. Turnbull: 9780814704196: Amazon.com: Books. For me, history was key to appreciating the motivation and insights of various mathematicians. E.g. logs were invented before the invention of function graphing. A rather long winded but (for me) enjoyable introduction with some elementary examples with a computer science emphasis is Gödel, Escher, Bach The US generally introduces proof by way of geometry. A nice, formal but readable treatment is in Oliver Byrne: The First Six Books of the Elements of Euclid: Werner Oechslin: 9783836544719: Amazon.com: Books | 11/19/2014 |
How can one evaluate and compare various applications of artificial intelligence? | Financially (e.g. cash flow, revenue, ROI, etc.) | 2/22/2017 |
How can one use random matrices in machine learning? | A neural network is usually initialized to random values before applying gradient descent algorithms. | 12/28/2017 |
How can we establish safe objectives to ASI (artificial super intelligence) so that they don't turn against us? | Create an AI predator | 1/2/2017 |
How can you be sure you are making the right career moves? | Robert Taylor is correct; but given that… The way I made the decision was to pick the job that trained me to do what I wanted to do in a few years. In my case, I passed up a bigger salary in exchange for exposure to how a startup worked. That approach worked well for me. | 8/11/2016 |
How can you increase sales and foot traffic to something like a gum ball machine? | Make it a poke stop and fill it w plastic pokeman characters or gum balls that look like poke balls | 8/4/2016 |
How could DeepMind be used for Calico? | Here's one: Large-Scale Machine Learning for Drug Discovery | 4/19/2016 |
How did you get into machine learning and what avenues do you suggest? Was this all exclusively in Norway? | 4/25/2018 | |
How did you learn about engineering? | By building stuff I was interested in. At the time (50 years ago) I lived in Houston, so I figured out how to make kites that I flew during hurricanes that could pull me very fast on my wagon. As I got older it was stuff like blue boxes, red boxes and black boxes (Blue box - Wikipedia). Then programs. In software, here are examples of how i figured out to make stuff fast (Bill Paseman's answer to Every time I begin a personal project with an idea, I fail at continuing it. What can I do so I won't change ideas fast?) Now I’m working on stuff using jet engines. For what its worth, I recently heard a Cornell MechE professor advise students to work in a bike shop. | 3/7/2018 |
How do I carefully share my idea? | There are several ways: patent, trademark, NDA, but they all boil down to mechanisms that give you the right to sue if the idea is stolen; none that actually keep the idea from being stolen. If the idea is a business: e.g. Mark Zuckerberg’s theft of the Winkelvoss Facebook idea, it is difficult collect even if the idea is stolen, there are witnesses and a contract is in place. Legally: an NDA is the simplest; although no VC or Angel in their right mind will sign one. (I believe) the most legally effective is to file a provisional patent. Record the date it is received by the USPTO and then disclose the idea. Anyone who replicates the idea after that date has violated the patent. However, you need to make sure the idea is patentable first. Realistically, (I believe) that the best way is to share it with someone you have known for a long time, trust, wishes you well and wants to see you succeed. | 7/20/2016 |
How do I create a startup similar to Deepmind? | My first startup was created around a product I loved. It failed. My second was created around a product that customers would pay dearly for. It succeeded. If you can tie your technology to a problem that people will pay to solve, I suspect it will have a better chance of success. | 5/1/2016 |
How do I do my science fair project? | How old are you and what subjects interest you? | 1/24/2017 |
How do I do my science fair project? | 11/20/2016 | |
How do I find great co foudners to form a good management team for a great idea that is already vetted with investors? | Looks like an add-on to your last question, so I will answer this one. Ask the investors who vetted your plan for suggestions (even if they did not invest). Besides the usual background checks, I brought in my first 5 employees as consultants. If they worked out, I made them an offer. I have to go through 4 VP marketing candidates that way before I found one that worked. | 7/25/2016 |