Author Archives: Anand Sarwate

Reads from the first half of 2009

Posted on by

Some reads from the first half of this year, in no particular order…

American Karma (Sunil Bhatia) — A qualitative study of a professional South Asian community in New England. Bhatia explores issues like how home/work are separated, how South Asian identity is maintained, and the stresses faced by these corporate employees. Of particular interest was how many would take accent reduction classes to move up the corporate ladder and the ways in which they would justify or apologize for their co-workers’ tokenization of them. There was also a lot about how the families interacted via their children with the school district. I thought it was a worthwhile read for people who are interested in South Asian American studies, but it might be a bit jargon-laden for some.

Funny You Don’t Look Like One (Drew Hayden Taylor) — This was a collection of essays by Ojibway writer Drew Hayden Taylor, collated from several different publications. I lacked context for a lot of what he talked about, such as Oka (warning, Wikipedia article is highly contested) and the Akwesasne cigarette “smuggling” debate. He is an engaging writer, and I enjoyed reading this book — it spurred me to read a bit more about the context, and that’s always a good thing.

The Karma of Brown Folk (Vijay Prashad) — Unlike Amardeep, I still think this book has a lot to offer South Asian Americans in terms of contextualizing the ties between India and the US diaspora and the ties that should exists between South Asians and other people of color in the US. Prashad paints a rather dire picture of things, but I think what is most lacking in South Asian youth is critical thinking, and this book does a good job of questioning the sociopolitical underpinnings of South Asian American culture, especially among the professional diaspora. Maybe it’s not a great book to teach from, and for sure it it biased, but it’s a groundbreaking work, I think. It has aged a bit (I last read it around when it came out), but I think it’s still valuable.

Making Money (Terry Pratchett) — This was a Discworld novel, this time sending up the banking industry. It was topical given the current crisis, and I found it entertaining in its formulaic way…

Steppin’ on a Rainbow (Kinky Friedman) — A sort of gonzo mystery novel set in Hawaii and full of schlock pulpy native stereotypes. Avoid.

Love Medicine (Louise Erdrich) — I really enjoyed this multi-generational novel about an extended Ojibwe family. It was a bit difficult to get into at first, but it definitely hooked me. What got me was how Erdrich gets under the complicated ways in which people show their love for each other and how inexplicable actions can make sense with the proper context…

Never Let Me Go (Kazuo Ishiguro) — This novel came highly recommended but I have to say that I wasn’t as enthralled with it. Ishiguro wrote an engrossing pseudo-dystopian narrative but the complacency of the narrator, rather than being harrowing, was simply disappointing. It did remind me a bit of some of the chapters in Cloud Atlas, but in the end I felt like the novel failed to make me care, somehow. Which is sad, because I really should care about these people. Maybe it’s more of a judgement on me…

Fifth Business (Roberton Davies) — This was also a recommendation, and I liked it, although not as much as R. Fifth Business is a memoir of a school teacher who grew up in a small town in Canada, fought through WWI and has the scars to show it. Although I did find the narrator a bit tiresome at times, I did like the form of a life-long bildungsroman.

The Ghost Brigades (John Scalzi) — This is a sequel to Old Man’s War and I didn’t like it as much as the original. Scalzi is often called the modern Heinlein, and like Heinlein, I found it a bit repetitive and was not too keen on its politics, such as they were.

Unruly Immigrants (Monisha Das Gupta) — This is a study of alternative social/political/economic movements within the South Asian American community. In particular, she looks at feminist, queer, and labor groups. She uses the phrases “place makers” to describe their activities versus the “place taker” actions that often characterize the majority South Asian community. I liked that turn of phrase. The book relies a lot on her own experiences with some of the groups as well as extensive interviews. One thing that pops out is the complexity of relations between South Asians from Asia and from the Caribbean and Africa, between different economic class groups when trying to organize domestic workers, and gender differences in labor and queer groups. It’s definitely worth reading for those who are interested in activism in the South Asian American community.

ISIT 2009 : talks part three

Posted on by

This is the penultimate post on papers I saw at ISIT 2009 that I thought were interesting and on which I took some notes. I think I’m getting lazier and lazier with the note taking — I went to lots more talks than these, but I’m only writing about a random subset.

Existence and Construction of Capacity-Achieving Network Codes for Distributed Storage
Yunnan Wu

This paper looked at the distributed storage problem using the framework of network coding that Alex has worked on. The basic idea is that you have a bunch of drives redundantly storing some information (say n drives and you can query any k to reconstruct the data). You want to make it so that if any drive fails, it can be replaced by a new drive so that the new drive doesn’t have to download too much data to maintain the “k out of n” property. The problem can be translated into a network code on an infinite graph. This paper talked about how to achieve the optimal tradeoff between the bandwidth needed to repair the code and the storage efficiency/redundancy. The key thing is that even though the network graph is infinite, the network code can be constructed over a field whose size only depends on the number of active disks. Schwartz-Zippel reared its head again in the proof…

Achievability Results for Statistical Learning under Communication Constraints
Maxim Raginsky

This talk was on how to learn a classifier when the data-label pairs (X_i, Y_i) must be compressed. Max talked about two different scenarios, one in which you can use R bits per pair, and the other in which only the labels Y_i must be sent. He defined a rate-distortion-like function where the generalization error the classifier played the role of the distortion. In the first case it’s easier for the encoder to learn the classifier itself and send the classifier over — this takes 0 bits per sample, asympotically, and incurs no extra generalization error. In the second case he uses a kind of Wyner-Ziv-like scheme which is a bit more involved. Compression is not a thing many machine learning folks think about, so I think there’s a lot of interesting stuff in this direction.

On Feedback in Network Source Coding
Mayank Bakshi, Michelle Effros

This paper was about how feedback can help in distributed source coding. In some lossless and lossy problems, feedback can strictly increase the rate region. Some of these problems took the form of “helper” scenarios where the decoder wants one of the sources but gets coded side information from another source. They show that this scenario holds more generally, and so feedback helps in general in network source coding.

Feedback Communication over Individual Channels
Power Adaptive Feedback Communication over an Additive Individual Noise Sequence Channel
Yuval Lomnitz, Meir Feder

These talks were near and dear to my heart since they dealt with coding over channels which are basically unknown. The basic idea is that you know only X^n and Y^n, the inputs and outputs of the channel. The trick is then to define a corresponding “achievable rate.” They define an empirical mutual information and show that it is asymptotically achievable under some circumstances. The scheme is based on random coding plus maximum mutual information (MMI) decoding. When feedback is present they can do some rateless coding. There’s a full version on ArXiV.

Upper Bounds to Error Probability with Feedback
Barış Nakiboğlu, Lizhong Zheng

Baris gave a nice talk on his approach to modifying the Gallager exponent bounds to the case when feedback is available. The encoder constantly adapts to force the decoder into a decision. The upper bound is based on a one-step/channel-use reduction to show an exponentially decaying error probability. One nice thing he mentioned is that at rates below capacity we can get a better error exponent by not using the capacity-achieving input distribution. Even though I read the paper and Barış explained it to me very patiently, I still don’t quite get what is going on here. That’s probably because I never really worked on error exponents…

Rationing healthcare

Posted on by

A recurring link in my facebook feed today was to an article in the NY Times Magazine on rationing health care. It’s worth reading, but the this made me squirm a bit:

But even in emergency rooms, people without health insurance may receive less health care than those with insurance. Joseph Doyle, a professor of economics at the Sloan School of Management at M.I.T., studied the records of people in Wisconsin who were injured in severe automobile accidents and had no choice but to go to the hospital. He estimated that those who had no health insurance received 20 percent less care and had a death rate 37 percent higher than those with health insurance. This difference held up even when those without health insurance were compared with those without automobile insurance, and with those on Medicaid — groups with whom they share some characteristics that might affect treatment. The lack of insurance seems to be what caused the greater number of deaths.

Oh correlation/causation fallacy, I long for your demise. At least he said “seems.”

Hangwringing about conference blogging from Nature

Posted on by

Lav pointed out this article in Nature on concerns over blogging about talks at conferences. It contains gems such as:

“I could take pictures of every slide and it would be on the Internet within seconds.” — Lars Jensen

and

MacArthur’s comprehensive postings were read by many scientists but they irked journalists attending the meeting. The meeting rules stated that reporters had to seek permission from speakers before publishing material on their work…

and

This kind of direct-to-web exposure creates problems for many industrial and applied researchers. In the United States, patent applications must be filed within a year of any information becoming available to the public. The exact date of that ‘public disclosure’ used to be difficult to nail down, but no more, says Michael Natan, chief executive officer of Oxonica Materials, a nanotechnology company in Mountain View, California. In the Internet age, time-stamped photographs of a talk can let competitors know the exact minute a researcher presented a patentable result. Consequently, “people in industry will be much more circumspect about what they present in public”, he says.

So I know I don’t work on Science (with a capital S) and that a I’m not the most knowledgeable guy out there. I do know from talking to friends that there is sometime shady behavior involving scooping of other labs by stealing ideas and fast-tracking a paper, but this article is a bit too paranoid.

  • Industry is already circumspect about what they present in public. I don’t think blogging is going to make them any more paranoid — patent firms already hire PhD engineers to comb the conference proceedings and literature to prove ideas were disclosed publically or invented too early in order to limit the scope of patents.
  • Who the hell would take pictures of every slide? Lars Jensen himself thinks it’s ridiculous (see the comments on the article) and the reporter here is definitely ginning up the controversy.
  • Going to a conference and talking publicly about your research is public disclosure. Sorry dudes, but we should not indulge in Clintonian verbal acrobatics.
  • If Cold Spring Harbor wants to force journalists to abide by ridiculous disclosure rules, then they should do what MILCOM does and have classified sessions.

Romanian diacritics

Posted on by

I came across this blog post today while trying to figure out how to write the Romanian breve (the symbol ă) in a document, and it was an amusingly angry rant about Romanian orthography. The fact that the Romanian currency even got it wrong is pretty funny. But it seems a bit like a futile battle; things always change and I bet the orthography gets merged eventually. I, for one, miss the ess-zett (ß) in German, but it’s gone the way of the dinosaurs.

That would be a great name for an diacritic mark — a dinosaur. A stegosaurus sitting on top of a U. But how would it be pronounced?

ISIT : first set of talks

Posted on by

Outer Bounds for User Cooperation
Ravi Tandon, Sennur Ulukus

This paper looked converse bounds for the MAC with generalized feedback. By applying the dependence balance bound, one can get an outer bound for the capacity region in terms of auxiliary random variables whose cardinality is difficult to bound. For a specific form of the channel, the “user cooperation channel.” where the feedback signals to the users are made from noisy versions of the other users’s signal, they find a way to evaluate the bound by looking at all input densities satisfying the constraints and then arguing that it is sufficient to consider Gaussian densities. The resulting bound matches the full-cooperation and no-cooperation bound when taking the respective limits in the feedback noise, and is tighter than the cutset bound.

Optimal Quantization of Random Measurements in Compressed Sensing
John Sun, Vivek Goyal

The goal here was to design a quantizer for noisy measurements in compressed sensing, while keeping the decoder/reconstructor fixed. This was done in the same high rate scalar quantization setting that I’d seen in other talks, with point-density functions to represent the reconstruction points. They draw a connection to the earlier work on functional scalar quantization and find optimal quantizers for the Lasso. They used an recent called “homotopy continuation,” which looks at the solution produced by the Lasso as a function of the regularization parameter, which I should probably read up on a bit more…

A Sparsity Detection Framework for On-Off Random Access Channels
Sundeep Rangan, Alyson Fletcher, Vivek Goyal

This looked at a MAC with n users and the decoder wants to detect which users transmitted. The users are each “on” with some fixed probability, and they show this can be put into a sparsity detection framework. There are two problems to overcome — the near-far effect of large dynamic range of the received signals, and a “multiaccess interference” (MAI) phenomenon that plagues most detection algorithms, including the Lasso and orthogonal matching pursuit (OMP). The question is whether a practical algorithm can overcome these effects — they show that a modification of OMP can do so as long as the decoder knows some additional information about the received signals, namely the power profile which is the order of the users’ power, conditional that they are active.

Eigen-Beamforming with Delayed Feedback and Channel Prediction
Tr Ramya, Srikrishna Bhashyam

This paper looked at the effect of delay in the feedback link in a system that is trying to do beamforming. The main effect of delay is that there is a mismatch between CSIT and CSIR. There is also the effect of channel estimation error. One solution is to do some channel prediction at the encoder, and they show that delayed feedback can affect the diversity order, while prediction can help improve the performance. Unfortunately, at high Doppler shift and and high SNR, the prediction filter becomes too complex. I hadn’t really seen much work on the effect of longer delay in feedback, so this was an interesting set of ideas for me.