Top Menu

Custom Search

Thursday, November 14, 2013

पाहून तुला तिची मला आठवण येते

त्याला तिच्या सारखी एक मुलगी दिसते. तिच्याकडे तो पाहताच राहतो. ती थोडी भांबावते. तो तिला या काही ओळींनी या सगळ्याच कारण सांगतो

नको असं - तसं काही  वाटून तू घेऊ
चेहऱ्यात तुझ्या तिचे गाल मऊ मऊ

डोळे जसे तुझे हसती हसायची ती ही
हसत हसत पाहून तिला फसायचो मी ही

डोळ्या मध्ये मला तुझ्या पाहूदेना थोडे
हळू हळू मग मी हे सोडवेल कोडे

पाहून तुला तिची मला आठवण येते
डोळ्यामध्ये  मग ती तरळून जाते

भेटलीच तुला तर आवर्जून सांग
कधी कधी फुटतो माझ्या डोळ्यांचा ही बांध

जमलेच तुला तर भेटतजा कधी
सुकलेल्या मनाला झुळूक ही साधी

                                                 - प्र. बा.

Sunday, August 25, 2013

लग्ना नंतरचे प्रेम ...

डॉक्टर ने त्याला सांगितलेले असते काही दिवस एका कुशीवर झोपू नका. ती रात्र भर जागीच असते मधे मधे त्याला एका कुशीवरून पाठीवर करत.
यालाच तर लग्ना नंतरचे प्रेम म्हणतात ना ?

Tuesday, July 30, 2013

NLP resources from Uni. of Edinburgh


Machine Learning for Natural Language Processing (ML-for-NLP)

This reading group focuses on Machine Learning techniques that may be applied to the field of Natural Language Processing. Participants are encouraged to suggest topics, papers, or tutorials (which need not involve any current application in NLP) by adding them to the lists below. Suggesting a paper does not constitute any sort of commitment to presenting that paper.
Meetings are approximately every week on Thursdays. Meetings will be in 4.02 at 3pm unless otherwise stated.
Announcements for this group will be made by email, and it is possible to sign up to the mailing list here.


  • No news, is good news

Tools for Research. Meeting notes

Unix Tools
The screen command. Cheatsheet, John's screen resources (John says: Note that the screen configuration files are named .screenrc and .screenrc.gen, and so you have to explicitly look for dotfiles to see them. The and scripts are pretty specific to my directory arrangement and may need tweaking for others' setups.)
Unix for Poets by Ken Church is a nice guide to using the unix tool set: tr,grep,sort,uniq,wc,rev,sed,awk,shuf,cat,tac,tail,cut,paste,etc.
Basic stuff but super useful
LaTeX & PDFs & Bibliography Management
JabRef - bibliography management
Okular - pdf highlighting
Research organization
Large-Scale Computation
Machine learning & Coding libraries



Paper Recommendations

Please add papers you consider appropriate for ML-for-NLP. Please also add thematic categories that are not covered.
Statistical Significance Testing
Loss Minimization
Bayesian Methods
Topic Models
Sampling Methods
Dynamical Systems
Information Theory
Variational Methods
Probabilistic Generative Models
Graphical Models
Deep learning and Energy Based Models
Language Modeling

Other Reading Groups

Useful Links

Previous meetings

Past meetings

3pm, Thursday 6th September, room 1.16 (Planning Meeting)

  • Discuss ideas for topics to be presented
  • Discuss whether or not you are happy for our meetings to become more exercise-based

3pm, Thursday 13th September, Room 3.02 (Information Theory Primer)

3pm, Thursday 20th September, Room 4.02 (Information Theory MacKay )

  • Chapter 2 of MacKay (pages 34-48 of the pdf, pages 22 -- 36 of the book), and focus on solutions:
    • Generation and inference:
      • Exercise 2.4
      • Exercise 2.5
      • Example 2.6

3pm, Thursday 27th September, room 4.02 (Information Theory MacKay )

  • Chapter 2 of MacKay (pages 34-48 of the pdf, pages 22 -- 36 of the book), and focus on solutions:
    • Bayesian predictive distribution:
      • Exercise 2.8
    • Jensen's inequality (important for all versions of EM):
      • Exercise 2.14
      • Example 2.15

3pm, Thursday 4th October, room 4.02 (Hypothesis Testing)

3pm, Thursday 18th October, room 4.02 (Variational EM)

bkj-VBwalkthrough.pdf: Variational EM

Meetings in 2011

Good advice on managing the paper presentation slides and talk


Some tips for preparing your paper presentation

First, organize your talk:

  1. Read the entire paper at least 3 times. 
    You need to be able to explain the details in the paper (even the ugly tricky notation)
    You need to be able to provide a critical analysis of the paper
    Check out references in the related work section of the paper. (this will help you put the paper in context of a larger body of work and will help you critique the paper's results/contributions)
    Look at Paper Reading Advice for more details.
  2. Find the important ideas A paper has many details but only one or two main ideas; structure your talk around these main ideas.
  3. Create a Talk Outline 
    Your talk should be organized in a top-down manner.
    You should have the following main sections in your talk:
    • Introduction, The Big Picture: what, why, how, and why we should care (motivation). Be sure to include:
      • a statement of the problem being solved (what)
      • motivation and putting the work in context (why and why should we care)
      • a high-level view of the author's solution (how)
    • Details of solution
    • Results demonstrating/proving their solution
    • Critic of Work (possibly compare to related work)
    • Conclusions & Future Directions for this work

    The talk should be organized as the important ideas first, the details second, conclusions last. Each section of your talk should be organized in a similar manor: high-level important points first, details second, summarize high-level points last. If the paper is well written, you can use the paper's organization as a guide.

Next, Design your slides

  1. Slide Organization Your slides should be organized like an outline--a few main points, with sub points under each one.
    Your slides are a guide for your talk not a word-for-word copy of your talk. List specific points that you want to talk about as sub-topics of each main topic. If there are particular details that you want to discuss, outline them on the slide and keep written notes for you to refer to in your talk rather than writing all the details on the slide.
  2. Summarize Main Points You should have a summary slide of the main ideas at the end.
    If applicable, Include a list of open questions from the paper
  3. It is okay to waste space Add just enough prose prose to present the main points and highlight the main parts of each point. Use phrases rather than complete sentences and use large fonts. You can use acronyms and abbreviations sparingly, however you should say the complete name when you talk about about them. For example, if you abbreviate processes to procs on a slide, say "processes" when you talk about the point not "procs". Similarly, if your create an acronym for your super fast multi-cast implementation SFMC and refer to the old slow multi-cast implementation as OSMC, then say "our super fast multi-cast" and "the old slow multi-cast" rather than "SFMC" and "OSMC". The exception is for well-known acronyms such as PVM, MPI, API, JVM, etc.
  4. A picture is worth a thousand words Use figures and graphs to explain implementation and results. It is very hard to describe a system implementation without having a picture of the components of the system. I once attended a talk about Intel's I64 architecture where the speaker tried to discuss the details of the layout of the chip and the interactions between the components without having any figures. It made for a very bad talk and a very hostile audience.
  5. Number of Slides As a general rule, it should take 2-3 minutes to talk through the material on one slide, so for a 45 minute talk you should have about 20 slides.If there is too much material in a paper to present completely in 45 minutes, then pick one part (the most interesting/important part) that you will discuss in detail, and present the other parts at a higher level. You can create back-up slides for specific details that you don't plan to talk about, but may get questions about.

Next, preparing your presentation

  1. Provide a talk road-map Tell audience where you are going with your talk.
    • Give audience a road-map of your talk at the beginning by using outline slides
      Immediately after the title slide, put up an outline slide and tell the audience the main organization of your talk. Another alternative is to first have a few slides motivating the paper's general topic, then put up an outline slide giving the audience a road-map of your talk.
    • It should be clear when you start a new high-level part of your talk
      Use good transitions from one slide to the next, and from one main topic to the next..."We just talked about the implementation of foo now we will look at how well foo performs for synthetic and real workloads.
      You may want to use the outline slide at other points in your talk to provide a visual transition between parts.
  2. Repeat Your Point There is a rule that says you have to tell your audience something three times before the really hear it:
    1. Tell them what you are going to say.
    2. Say it.
    3. Summarize what you said.
    This is particularly important for figures and graphs. For example:
    1. This graph show how the A algorithm performs better than the B and C algorithms as the number of nodes increase
    2. The X axis is number of nodes, the Y axis is execution time in seconds The red curve shows the execution time of A as the number of nodes increases The blue curve shows ...
    3. Thus you can see that as the number of nodes increases above N, the A algorithm performs better. This is because of increased message traffic in algorithms B and C as shown on the next slide...
    4. Explain concepts in your own words It is certainly okay to lift key phrases from the paper to use in your talk. However, you should also try to summarize the main ideas of the paper in your own words.
    5. Talk to the Audience Don't read your slide off the screen, nor directly off the projector. It is okay to stop for a second and refer to your notes if you need to.
    6. Practice Give a practice run-through of your talk. Stand in a room for 1 hour and talk through all your slides (out loud). This should be a timed dress rehearsal (don't stop and fix slides as you go). Members of your reading group should provide a practice audience for you.
    7. Nervousness: How to fight back
      • A well organized, practiced talk will almost always go well. If you draw a blank, then looking at your slides will help you get back on track.
      • Taking a deep breath will clam you down. One trick is to try to remember to take a deep breath between each slide.
      • Slow down. Take a few seconds to think about a question that is being asked before you answer it. It is okay to pause for a few seconds between points and between slides; a second or two of silence between points is noticeable only to you, but if you are talking a mile a minute everyone will notice.
      • Bring notes. if you are afraid that you will forget a point or will forget your elegant transition between slides 11 and 12, write these down on a piece of paper and bring it with you. However, you don't want to have a verbatim copy of your talk, instead write down key phrases that you want to remember to say.
      • Give at least one practice talk to an audience.
      • Be prepared to answer questions. You don't have to know the answer to every question, however you should be prepared to answer questions and able to answer most questions about the paper. Before you give the talk, think about what questions you are likely to get, and how you would answer them. You may want to have back-up slides ready for answering certain questions.
      • It is okay to say "I don't know" or better yet "gee, I hadn't thought about that, but one possible approach would be to..." or to refer to your notes to answer questions.
    8. Talk to me...this is not optional
      You should meet with me the Tuesday before your talk with your talk outline in hand, and the Thursday or Friday before your talk with your talk slides.

Wednesday, May 15, 2013

Getting Better- Dharmesh Shah on Linkedin


Advice is often much easier to give than to take. For example, I was at a meeting at HubSpot yesterday and we were talking about a particularly area of the business that is having some challenges. A plan was presented. Then, a question came up in the room: “What you're suggesting we do is not that different from what we've done in the past. Why do we think it's going to lead to a dramatically different outcome?” It was a good question.
This just doesn't just apply to businesses, it applies to individuals as well. When I’m facing a problem or challenge, one of the key questions is to ask: “Okay. So what do I plan to do differently so I can have a chance at a different outcome?”
The only way to overcome a problem is to do something differently because unless you do, tomorrow is just like today. (And who wants that?)
But there's an even better approach: Instead of waiting until you have to make a bad situation better, why not turn an “okay” situation into a great one by tackling a challenge head on?
Here are a number of ways you can make significant improvements by doing things differently:
1. Take the opposite approach.
If you’re struggling to reach a goal then clearly what you’re currently doing isn't working. Think about the overall result you want to achieve, and instead of continuing to tweak and refine your approach try tackling it from an entirely different direction. (For fun, see “Startup Advice From George Costanza”)
For example, if you need more sales and conventional advertising approaches aren’t working, try writing a blog or interacting on social media. If you are finding it hard to convince your colleagues on a radical new idea or project, try ditching the PowerPoint slides and writing a detailed document in long-form. If you want to lose weight but diets don’t work, decide you’ll train for and run a marathon. (You may see a heavy person run, but you’ll never see a heavy runner.)
Pick one goal you're struggling to achieve and try a completely different approach. While sometimes making small adjustments eventually pays off, occasionally you just need to blow things up and start fresh.
2. Eliminate one goal.
We love to set goals. So we often set too many goals, and it's impossible to do ten or even five things incredibly well.
Take a look at your list of goals and pick at least one that you will consciously set aside, at least for now. Don't feel like you’re giving something up. If you weren’t accomplishing your goals anyway, you’re really not losing anything.
Then put the time you were spending on that goal into achieving your highest priority goal.
We can't have it all, but we can have a lot, especially when we narrow our focus to achieving one or two of our most important goals.
3. Modify your routine.
Get up earlier. Or stay up later. Check in with key employees before you plop down at your desk. Eat at your desk. Eat with a different group every day. Shuffle the items on your daily to-do list. In short, shake up what you regularly do.
For example, at HubSpot a large part of the company goes through a random “seat shuffle” every three months. It reflects our credo that change is constant and definitely shakes up the status quo.
Pick one thing you do on a regular basis, preferably something you do for no better reason than that's the way you’ve always done it, and do it in a different way or at a different time.
Familiarity tends to breeds complacency, and complacency is a progress killer.
4. Add a new metric.
Other people measure your performance, but we all have ways we measure our own performance as well. Maybe you focus more on quality, or on the time it takes to complete a task, or on customer satisfaction, or….
Whatever method you use is surely effective, but constantly using only one or two measurements could cause you to miss opportunities to improve.
Say you’re in sales and you (understandably) focus on meeting your sales goals. But, for a week, try focusing less on how many customers you close and more on the quality of the conversations you have.
Measuring your performance in a different way requires you to look at what you do from a new perspective. Keep your current measurements, then pick a process and measure your performance in a different way for a period of time.
I promise you’ll find ways to do it even better.
5. Help a colleague.
Most people won’t ask for help. So don’t wait: Pick someone who is struggling and offer to help.
Yet don't simply say, "Is there some way I can help you?" Be specific. Offer to help with a specific function. Offer to take over a task for a few days so they can catch up on other work. Offer to pitch in on the spot and work together.
A generic offer is easy to brush aside, especially since most of us hate to admit when we need help. A specific offer shows not just that you want to help but also that you care.
When you help others succeed you automatically succeed too.
6. Help a superstar.
Compared to other employees the best performers don't tend to need help. So they rarely get it.
Ask if you can help a superstar perform a specific task. You’ll definitely learn something. At least a few of her skills, attitudes, and techniques will rub off on you.
7. Just help.
Few things feel better than helping someone in need. Take a quick look around; people less fortunate than you are everywhere. It takes very little time and effort to make a meaningful difference.
And not only will you make a difference in someone else’s life, you’ll also make a difference in yours. The better you feel about yourself, the more enthusiastic and motivated you will be.
We can all use more of that.
8. Adopt a habit.
Successful people are successful for a reason, and that reason is often due to the habits they create and maintain.
Take a close look at the people who are successful in your field: What do they consistently do? Then adopt one of their habits and make it your own.
Never reinvent a wheel when a great wheel already exists.
9. Embrace others as they "are".
The company you work for probably isn't going to change. The person you report to probably isn't going to change. Your suppliers, your vendors, your customers…. they aren't going to change.
Stop expecting them to. When you stop focusing on negatives you may start to notice positive qualities you missed. (No one is as terrible or as great as you assume.)
Pick one source of frustration and decide what you will do differently. You can’t change other people, but you can always change yourself – and in the process make things better for everyone.
11. Embrace who you “are”.
I would love to be as funny as him. Or be as smart as him. Or make an impact like him.
That’s not likely to happen.
And mostly I'm okay with that because I can always be a better version of me. I can learn to be funnier. I can keep trying to know more. I can always do more to make a bigger difference in the lives of our employees and my family and friends.
Think about the people you admire. Then, instead of focusing on their accomplishments, pick a few of their admirable qualities you want to emulate.
You can never be the people you admire, and the great thing is they can never be you. Embrace who you are and keep trying to be a better version of who you are.
That is all any of us can do.

Sunday, March 17, 2013

Important career and life advice


Sometimes the route to happiness depends more on what you don't do.
sad and happy smiley face cupcakes
Flickr Creative Commons

Happiness--in your business life and your personal life--is often a matter of subtraction, not addition.
Consider, for example, what happens when you stop doing the following 10 things:
1. Blaming.
People make mistakes. Employees don't meet your expectations. Vendors don't deliver on time.
So you blame them for your problems.
But you're also to blame. Maybe you didn't provide enough training. Maybe you didn't build in enough of a buffer. Maybe you asked too much, too soon.
Taking responsibility when things go wrong instead of blaming others isn't masochistic, it's empowering--because then you focus on doing things better or smarter next time.
And when you get better or smarter, you also get happier.
2. Impressing.
No one likes you for your clothes, your car, your possessions, your title, or your accomplishments. Those are all "things." People may like your things--but that doesn't mean they like you.
Sure, superficially they might seem to, but superficial is also insubstantial, and a relationship that is not based on substance is not a real relationship.
Genuine relationships make you happier, and you'll only form genuine relationships when you stop trying to impress and start trying to just be yourself.
3. Clinging.
When you're afraid or insecure, you hold on tightly to what you know, even if what you know isn't particularly good for you.
An absence of fear or insecurity isn't happiness: It's just an absence of fear or insecurity.
Holding on to what you think you need won't make you happier; letting go so you can reach for and try to earn what you want will.
Even if you don't succeed in earning what you want, the act of trying alone will make you feel better about yourself.
4. Interrupting.
Interrupting isn't just rude. When you interrupt someone, what you're really saying is, "I'm not listening to you so I can understand what you're saying; I'm listening to you so I can decide what I want to say."
Want people to like you? Listen to what they say. Focus on what they say. Ask questions to make sure you understand what they say.
They'll love you for it--and you'll love how that makes you feel.
5. Whining.
Your words have power, especially over you. Whining about your problems makes you feel worse, not better.
If something is wrong, don't waste time complaining. Put that effort into making the situation better. Unless you want to whine about it forever, eventually you'll have to do that. So why waste time? Fix it now.
Don't talk about what's wrong. Talk about how you'll make things better, even if that conversation is only with yourself.
And do the same with your friends or colleagues. Don't just be the shoulder they cry on.
Friends don't let friends whine--friends help friends make their lives better.
6. Controlling.
Yeah, you're the boss. Yeah, you're the titan of industry. Yeah, you're the small tail that wags a huge dog.
Still, the only thing you really control is you. If you find yourself trying hard to control other people, you've decided that you, your goals, your dreams, or even just your opinions are more important than theirs.
Plus, control is short term at best, because it often requires force, or fear, or authority, or some form of pressure--none of those let you feel good about yourself.
Find people who want to go where you're going. They'll work harder, have more fun, and create better business and personal relationships.
And all of you will be happier.
7. Criticizing.
Yeah, you're more educated. Yeah, you're more experienced. Yeah, you've been around more blocks and climbed more mountains and slayed more dragons.
That doesn't make you smarter, or better, or more insightful.
That just makes you you: unique, matchless, one of a kind, but in the end, just you.
Just like everyone else--including your employees.
Everyone is different: not better, not worse, just different. Appreciate the differences instead of the shortcomings and you'll see people--and yourself--in a better light.
8. Preaching.
Criticizing has a brother. His name is Preaching. They share the same father: Judging.
The higher you rise and the more you accomplish, the more likely you are to think you know everything--and to tell people everything you think you know.
When you speak with more finality than foundation, people may hear you but they don't listen. Few things are sadder and leave you feeling less happy.
9. Dwelling.
The past is valuable. Learn from your mistakes. Learn from the mistakes of others.
Then let it go.
Easier said than done? It depends on your focus. When something bad happens to you, see that as a chance to learn something you didn't know. When another person makes a mistake, see that as an opportunity to be kind, forgiving, and understanding.
The past is just training; it doesn't define you. Think about what went wrong, but only in terms of how you will make sure that, next time, you and the people around you will know how to make sure it goes right.
10. Fearing.
We're all afraid: of what might or might not happen, of what we can't change, or what we won't be able to do, or how other people might perceive us.
So it's easier to hesitate, to wait for the right moment, to decide we need to think a little longer or do some more research or explore a few more alternatives.
Meanwhile days, weeks, months, and even years pass us by.
And so do our dreams.
Don't let your fears hold you back. Whatever you've been planning, whatever you've imagined, whatever you've dreamed of, get started on it today.
If you want to start a business, take the first step. If you want to change careers, take the first step. If you want to expand or enter a new market or offer new products or services, take the first step.
Put your fears aside and get started. Do something. Do anything.
Otherwise, today is gone. Once tomorrow comes, today is lost forever.
Today is the most precious asset you own--and is the one thing you should truly fear wasting.

Two very Important take away from the article on in general life and professional journey 
  • Everyone is different: not better, not worse, just different.
  • The past is just training; it doesn't define you. 

Tuesday, January 15, 2013


I don't want just the fight of words, I want ideas to collide for creation of better ones. I don't want just the arrows of definitions and illustrations to be shot for nothing of any worth. Everything should have a meaning, a meaning that is intended.