Received this morning:

Dear Dr. Bob L. Sturm,

The international conference: 2012 3rd International Conference on Legal Medicine, Medical Negligence and Litigation in Medical Practice & 3rd International Conference on Current Trends in Forensic Sciences, Forensic Medicine & Toxicology will be held on 3rd to 5th February, 2012 in Jaipur, Rajasthan, India. This event will be organized and hosted by Saraswathi Institute of Medical Sciences, Ghaziabad, Uttar Pradesh, India. EPS Inc., a Canada-based biomedical consulting agency, is authorized by the Scientific Committee to co-organize this important and exciting educational event.

On behalf of the organizing committee, we cordially invite you to attend the conferences to present your recent work and ideas and share your knowledge in these specific fields. For the conference program details, please visit our websites at www.epsglobal.ca and www.epsworldlink.com.

We are very interested in your article **Recursive nearest neighbor search in a sparse and multiscale domain for comparing audio signals** that is published on **Signal Processing**. This article includes some novel conceptions, which may impress the worldwide experts in your field. The conferences will present an excellent opportunity for you to introduce this article to the worldwide experts, highlighting the great significance of your research achievement. We believe you will enjoy the casual and interactive settings of this meeting...

---

Of course I will skip this opportunity. I just wish I had an automatic reply to these things:

Dear **2012 3rd International Conference on Legal Medicine, Medical Negligence
and Litigation in Medical Practice & 3rd International Conference on
Current Trends in Forensic Sciences, Forensic Medicine & Toxicology**,

I appreciate your enthusiastic and bold invitation to present my work at your conference in **Jaipur, Rajasthan, India**, but I have already received a similar request from the **5th International Multi-Conference on Engineering and Technological Innovation: IMETI 2012**. Their transrobotic agent emailed me first, and I don't think it polite for me to decline, especially since their form email was thoughtfully in color.

Please consider me again in the future, especially after I obtain tenure.

Sincerely,

*Over the next few days, I am posting portions of a paper I am currently writing about making greedy algorithms as efficient as possible. I post this material before submission because: 1) much of it has been discussed before in a variety of places, and I am merely assembling some of these methods in one place; 2) I think it can find immediate use, rather than wait to release the entire paper after the paper deadline in February. Today, we begin with some notation, and a look at the naïve implementation of orthogonal matching pursuit (OMP). This provides us a baseline complexity.*

We are interested in efficiently modeling a signal \(\vu\)
by a linear combination of the atoms defined in a dictionary
\(\mathcal{D} = \{ \vphi_\omega \in C^M \}_{\omega \in \Omega= \{1, 2, \ldots, N\}}\):
$$
\vu = \MPhi \vx + \vn
$$
where \(\MPhi := [\vphi_1 | \vphi_2 | \cdots | \vphi_N]\).
By \(\Omega_k \subset \Omega\) we denote
an ordered subset of cardinality \(k\) of indices into the dictionary.
We denote the \(i\)th element of a set by \(\mathcal{I}(i)\).
The matrix \(\MPhi_{\Omega_k}\) is thus composed of
the ordered atoms indexed by \(\Omega_k\).
We denote \(\MP_k\) the orthogonal projection matrix onto
the range space of \(\MPhi_{\Omega_k}\), or equivalently,
onto the span of the subdictionary \(\mathcal{D}_k \subset \mathcal{D}\).
The projection matrix \(\MP_k^\perp = \MI - \MP_k\) is thus the
orthogonal projection onto the subspace orthogonal to
the span of the subdictionary,
or equivalently, the left null space of \(\MPhi_{\Omega_k}\).
We denote the inner product between two vectors in \(C^M\)
as \(\langle \vu_1, \vu_2 \rangle = \vu_2^H \vu_1\),
where \(^H\) denotes the complex conjugate transpose.
The vector \(\ve_k\) is the \(k\)th element of the standard basis of \(R^N\),
i.e., all zeros with a 1 in its \(k\)th row.
Finally, all norms are implicitly the \(\ell_2\)-norm,
\(\|\vx\|^2 = \langle \vx, \vx \rangle\).