Ideas must change in response to feedback from the problem to be solved.
The initial idea and its nth iteration are downstream from the real problem. What matters, therefore, is the iterative execution of those solutions. So instead of basing our judgment on the quality of the idea, we should rely on the execution ability because that is fairly constant in the development loop.
If you end up on the far-side of an idea-driven development process, you'll have solutions looking for a problem.
We don't build software like a bridge; we grow it like a garden. The industrial revolution was about building machines that dominate its environment. The software revolution is about writing code to resonate with its environment. The software grows in harmony with its user and business needs. Except, of course, for special cases such as this.
"The whole approach to developing software is intentionally designed not to rely on any particular person."
Our programming culture is not ready for this. Any programmer who identifies with their craft will never want to be another cog in the machine. A programmer could have chosen to be a doctor, lawyer, engineer, etc. but they choose to code because it is one of the way they can "express" themselves.
The analogy of maps and boats is just an instance of the exploration-exploitation idea. I have seen instances of this pop up every time we discuss problem solving in some form. It falls in the perennial variety of ideas that can't be revisited enough.
The paper by Sproat and Jaitly which introduces the challenge rightly notes that the acceptability of errors and quality of output is more important than accuracy for a real application. The number of instances in all of the critical semiotic classes is too low (1-2k for some, even less than 100 for others) for a meaningful comparison in accuracy.
But you are right to point out that the 'unacceptability' of errors could be analyzed better. However, we could not think of a way to quantify or form a metric that measures such errors. These 'silly' errors are subjective by their very nature and depend on a human reading them. As you have suggested, we are working on preparing a table of sorts to summarize all these errors and show a link between the availability/frequency of particular types examples to the performance of our model on those particular types. Something of this sort for example:
* The training set had 17,712 examples in DATE of the form xx/yy/zzzz. Upon the analyzing the mistakes in DATE class we did not find any mistakes made in the dates of the above form.
* On the other hand, if look into the mistakes made in MEASURE class we find that the DNC network made exactly 4 mistakes. The mistakes were reported in the units (g/cm3, ch, mA). Upon searching for the occurrences in the training set of these units, we found out that 'mA' occurs 3 three times, 'g/cm3' occurred 7 times and 'ch' occurred 8 times, whereas other measurement units like kg occur 296 times and cm occur 600+ times.
If you have any other ideas on how to analyze and report the results, please let us know. We will be glad to improve the quality of our work (By the way, we are undergrads and this is our very first research paper). Thanks again!
Thanks.
I'm currently taking the OS course this sem and we plan to build a minimal OS for raspberry pi from scratch using the linux kernel, writing system programs on top of it.
For internships, I believe start ups are the best bet to gain experience and maybe gain a mentor for life.
Start-up internships and GSOC are the best use of a summer vacation, I agree.
And regarding software education, I believe MOOC is going to play a very big role in the near future especially in countries like India where hard working and deserving students lack quality world class education.
And given the Indian affinity for certificates and credentials, I believe this could be a great start up idea too.
I love to talk to other passionate CS students too. Helps me reorient and correct my perspective.
And I'm gonna follow your advice and blog about my learning experiences. Infact, there are valuable gems of advice here and I'm thinking to compile them together in a blog post for many other curious students like me.
Being a good programmer and having a good GPA seems like inversely proportional in our college. Only the students who follow lecture slides and are good at writing exams seem to have >=9 CGPA.
Indeed.
After talking to other people and thinking hard about my options, I have come to same conclusion.
But finding good people to work with or under seems really hard. I really don't know how to find good mentors other than contributing to open source and hoping someone senior likes my contributions.
College is a good place to try things out before diving into them. But I have found no passionate profs in my college and getting a research internship under a good prof is a task in itself.
Though I have an interest in games and AI / ML, I would still consider being a good software engineer first. I'm aiming for a strong foundation and expertise in one field: AI or anything related. Being a fresher, I believe I won't get to work straight at these.
And by "at par", I meant the same opportunities that a top school student gets.
Most of my undergrad friends are really torn between academia and industry. And since I'm in my sophomore year, I feel it's too early to make the right call.
Anyways, I do need to decide sooner or later.
Yeah exactly but I always feel like I'm short of time. Like just a few years more and I'll have to look for a job. Mandatory attendance at boring classes doesn't help much too.
The initial idea and its nth iteration are downstream from the real problem. What matters, therefore, is the iterative execution of those solutions. So instead of basing our judgment on the quality of the idea, we should rely on the execution ability because that is fairly constant in the development loop.
If you end up on the far-side of an idea-driven development process, you'll have solutions looking for a problem.