DOCSLIB.ORG

Applications of the Shannon-Hartley Theorem to Data Streams and Sparse Recovery

Total Page：16

File Type：pdf, Size：1020Kb

Download full-text PDF Read full-text
Applications of the Shannon-Hartley Theorem to Data Streams and Sparse Recovery Applications of the Shannon-Hartley Theorem to Data Streams and Sparse Recovery Eric Price David P. Woodruff MIT IBM Almaden Abstract—The Shannon-Hartley theorem bounds the is also needed for network aggregation, which usually maximum rate at which information can be transmitted follows a bottom-up approach [18]: given a routing tree over a Gaussian channel in terms of the ratio of the signal where the nodes represent sensors, starting from the to noise power. We show two unexpected applications of this theorem in computer science: (1) we give a much simpler leaves the aggregation propagates upwards to the root. proof of an Ω(n1−2=p) bound on the number of linear We refer to the rows of A as measurements. measurements required to approximate the p-th frequency Alon, Matias, and Szegedy [2] initiated the line of moment in a data stream, and show a new distribution work on frequency moments. There is a long line of which is hard for this problem, (2) we show that the number of measurements needed to solve the k-sparse upper bounds on the number of linear measurements; we recovery problem on an n-dimensional vector x with the refer the reader to the most recent works [3], [11] and the C-approximate `2=`2 guarantee is Ω(k log(n=k)= log C). references therein. Similarly, we refer the reader to the We complement this result with an almost matching ∗ most recent lower bounds [16], [22] and the references O(k log k log(n=k)= log C) upper bound. therein. The best upper and lower bounds for obtaining I. INTRODUCTION a (1 + )-approximation with probability at least 1 − δ have the form n1−2=p · poly(−1 log(nδ−1). Let S be a real-valued random variable with E[S2] = τ 2. Consider the random variable S + T , where T ∼ The existing lower bounds are rather involved, using N(0; σ2) is additive white Gaussian noise of variance the direct sum paradigm for information complexity σ2. The Shannon-Hartley theorem states that [5]. Moreover, they apply to the number of bits rather than the number of linear measurements, and typically 2 1 τ do not provide an explicit distribution which is hard. I(S; S + T ) ≤ log 1 + ; 2 σ2 These issues can be resolved using techniques from [4], [20] and [15]. The resulting hard distribution is: choose where I(X; Y ) = h(X) − h(XjY ) is the mu- x 2 {−1; 0; 1gn uniformly at random, and then with tual information between X and Y , and h(X) = probability 1=2, replace a random coordinate x of x − R f(x) log f(x)dx is the differential entropy of a i X with a value in Θ(n1=p). F (x) changes by a constant random variable X with probability density function f. p factor in the two cases, and so the approximation algo- We show two unexpected applications of the Shannon- rithm must determine which case we are in. Hartley theorem in computer science, the first to estimat- ing frequency moments in a data stream, and the second We instead consider the following continuous dis- to approximating a vector by a sparse vector. tribution: choose x to be a random N(0;In) vector, i.e., a vector whose coordinates are independent stan- A. Sketching Frequency Moments dard normal random variables. With probability 1=2, In the data stream literature, a line of work has consid- replace a random coordinate xi of x with a value 1=p ered the problem of estimating the frequency moments in Θ(n ). The use of Gaussians instead of signs p Pn p n allows us to derive our lower bound almost immedi- Fp(x) = kxkp = i=1 jxij , where x 2 R and p ≥ 2. One usually wants a linear sketch, that is, we choose a ately from the Shannon-Hartley theorem. We obtain an m×n Ω(n1−2=p) bound on the number of linear measurements random matrix A 2 R from a certain distribution, for m n, and compute Ax, from which one can required for estimating Fp, matching known bounds up to poly(−1 log(Mnδ−1) factors. Our proof is much output a constant-factor approximation to Fp(x) with high probability. Linearity is crucial for distributed com- simpler than previous proofs. putation, formalized in the MUD (Massive Unordered Our new hard distribution may also more accurately Distributed) model [9]. In this model the vector x is model those signals x arising in practice, since it corre- split into pieces x1; : : : ; xr, each of which is handled by sponds to a signal with support 1 which is corrupted by a different machine. The machines individually compute independent Gaussian noise in each coordinate. Identify- Ax1; : : : ; Axr, and an aggregation function combines ing natural hard distributions has been studied for other these to compute Ax and estimate Fp(x). Linearity data stream problems, see, e.g., [19] and [17]. 0 0 B. Sparse Recovery r ≥ 2Gpn, then Bob sets Z = 1, else Bob sets Z = 0. 0 In the field of compressed sensing, a standard problem We thus have a Markov chain `; Z ! x ! y ! Z . is that of stable sparse recovery: we want a distribution If A works for any x with probability 1 − δ, as a m×n n distribution over A, then there is a specific A and random A of matrices A 2 R such that, for any x 2 R and with probability 1 − δ > 2=3 over A 2 A, there is an seed such that A, together with the associated estimation algorithm to recover x^ from Ax with procedure, succeeds with probability 1−δ over x drawn from the distribution described above. Let us fix this 0 kx^ − xkp ≤ (1 + ) min kx − x kp choice of A and associated random seed, so that Alice k−sparse x0 and Bob run deterministic algorithms. Let m be the for some > 0 and norm p. We call this a (1 + )- number of rows of A. We can assume the rows of A are approximate `p=`p recovery scheme with failure proba- orthonormal since this can be done in post-processing. bility δ. We will focus on the popular case of p = 2. Lemma 2: I(Z; Z0) = O(m=n1−2=p). For any constant δ > 0 and any satisfying = O(1) Proof: Let the rows of A be denoted v1; : : : ; vm. −1=2 and = Ω(n ), the optimal number of measurements Then we have that is Θ(k log(n=k)/). The upper bound is in [12], and the i 1=p 1=p i 0 lower bound is given by [1], [6], [14], [20]; see [20] for yi = hv ; xi = (4Gp) n · hv ; e`iZ + wi; a comparison of these works. 0 1=p 1=p One question is if the number of measurements can where wi ∼ N(0; 1). Define zi = (4Gp) n · i 0 be improved when the approximation factor C = 1 + hv ; e`iZ so yi = zi + wi. Then is very large (i.e. !(1)). In the limiting case of C = 1, 2 1 2=p i 2 corresponding to sparse recovery in the absence of noise, EZ;`[z ] = · (4Gpn) E`[(v ) ] i 2 ` it is known that O(k) measurements are sufficient [7]. 1 (4G n)2=p 1 (4G )2=p However, the intermediate regime has not been well = p = p = Θ(1=n1−2=p): 2 n 2 1−2=p studied. n 0 Using the Shannon-Hartley theorem, we prove an Hence, yi = zi + wi is a Gaussian channel with power 2 1−2=p Ω(k log(n=k)= log C) lower bound on the number of constraint E[zi ] = Θ(1=n ) and noise variance 0 2 measurements. We complement this with a novel sparse E[(wi) ] = 1. By the Shannon-Hartley theorem, recovery algorithm, which builds upon [12] and [13], 2 but is the first to obtain an improved bound for C > 1. 1 E[zi ] max I(zi; yi) ≤ log 1 + ∗ i 0 2 Our bound is O(k + k log(n=k) log k= log C), which v 2 E[(wi) ] ∗ matches our lower bound up to a log k factor. Because 1 1−2=p 1−2=p 1 = log 1 + Θ(1=n ) = Θ(1=n ): log(1 + ) ≈ , these results match the Θ( k log(n=k)) 2 results for 1. By the data processing inequality for Markov chains and Related work. Related lower bounds have appeared the chain rule for entropy, in a number of recent works, including [6], [14], [1], [21], and [10]. See [20] for a comparison. I(Z; Z0) ≤ I(z; y) = h(y) − h(yjz) II. LOWER BOUND FOR FREQUENCY MOMENTS = h(y) − h(y − zjz) X 0 0 0 This section is devoted to proving the following the- = h(y) − h(wijz; w1; : : : ; wi−1) orem: i X 0 X 0 Theorem 1: Any sketching algorithm for Fp up to a = h(y) − h(wi) ≤ h(yi) − h(wi) factor of (1 ± ) for < 1=2, which succeeds with i i X X probability 1 − δ for a sufficiently small constant δ > 0, = h(yi) − h(yijzi) = I(yi; zi) 1−2=p requires m = Ω(n ). i i p 1−2=p Let Gp = E[jXj ] where X ∼ N(0; 1). For constant ≤ O(m=n ): p, Gp is Θ(1). Consider the following communication game between p two players, Alice and Bob. Alice chooses a random Proof of Theorem 1: If Z = 1, then kxkp ≥ 4Gp·n, ` 2 [n] and associated standard unit vector e` = and so any (1 ± )-approximation is at least 2Gpn for n p (0;:::; 0; 1; 0;:::; 0) 2 R .
Load more

Recommended publications
  • Guide for the Use of the International System of Units (SI)
    Guide for the Use of the International System of Units (SI) m kg s cd SI mol K A NIST Special Publication 811 2008 Edition Ambler Thompson and Barry N. Taylor NIST Special Publication 811 2008 Edition Guide for the Use of the International System of Units (SI) Ambler Thompson Technology Services and Barry N. Taylor Physics Laboratory National Institute of Standards and Technology Gaithersburg, MD 20899 (Supersedes NIST Special Publication 811, 1995 Edition, April 1995) March 2008 U.S. Department of Commerce Carlos M. Gutierrez, Secretary National Institute of Standards and Technology James M. Turner, Acting Director National Institute of Standards and Technology Special Publication 811, 2008 Edition (Supersedes NIST Special Publication 811, April 1995 Edition) Natl. Inst. Stand. Technol. Spec. Publ. 811, 2008 Ed., 85 pages (March 2008; 2nd printing November 2008) CODEN: NSPUE3 Note on 2nd printing: This 2nd printing dated November 2008 of NIST SP811 corrects a number of minor typographical errors present in the 1st printing dated March 2008. Guide for the Use of the International System of Units (SI) Preface The International System of Units, universally abbreviated SI (from the French Le Système International d’Unités), is the modern metric system of measurement. Long the dominant measurement system used in science, the SI is becoming the dominant measurement system used in international commerce. The Omnibus Trade and Competitiveness Act of August 1988 [Public Law (PL) 100-418] changed the name of the National Bureau of Standards (NBS) to the National Institute of Standards and Technology (NIST) and gave to NIST the added task of helping U.S.
    [Show full text]
  • On the Irresistible Efficiency of Signal Processing Methods in Quantum
    On the Irresistible Efficiency of Signal Processing Methods in Quantum Computing 1,2 2 Andreas Klappenecker∗ , Martin R¨otteler† 1Department of Computer Science, Texas A&M University, College Station, TX 77843-3112, USA 2 Institut f¨ur Algorithmen und Kognitive Systeme, Universit¨at Karlsruhe,‡ Am Fasanengarten 5, D-76 128 Karlsruhe, Germany February 1, 2008 Abstract We show that many well-known signal transforms allow highly ef- ficient realizations on a quantum computer. We explain some elemen- tary quantum circuits and review the construction of the Quantum Fourier Transform. We derive quantum circuits for the Discrete Co- sine and Sine Transforms, and for the Discrete Hartley transform. We show that at most O(log2 N) elementary quantum gates are necessary to implement any of those transforms for input sequences of length N. arXiv:quant-ph/0111039v1 7 Nov 2001 §1 Introduction Quantum computers have the potential to solve certain problems at much higher speed than any classical computer. Some evidence for this statement is given by Shor’s algorithm to factor integers in polynomial time on a quan- tum computer. A crucial part of Shor’s algorithm depends on the discrete Fourier transform. The time complexity of the quantum Fourier transform is polylogarithmic in the length of the input signal. It is natural to ask whether other signal transforms allow for similar speed-ups. We briefly recall some properties of quantum circuits and construct the quantum Fourier transform. The main part of this paper is concerned with ∗e-mail: [email protected] †e-mail: [email protected] ‡research group Quantum Computing, Professor Thomas Beth the construction of quantum circuits for the discrete Cosine transforms, for the discrete Sine transforms, and for the discrete Hartley transform.
    [Show full text]
  • Bits, Bans Y Nats: Unidades De Medida De Cantidad De Información
    Bits, bans y nats: Unidades de medida de cantidad de información Junio de 2017 Apellidos, Nombre: Flores Asenjo, Santiago J. (sfl[email protected]) Departamento: Dep. de Comunicaciones Centro: EPS de Gandia 1 Introducción: bits Resumen Todo el mundo conoce el bit como unidad de medida de can- tidad de información, pero pocos utilizan otras unidades también válidas tanto para esta magnitud como para la entropía en el ám- bito de la Teoría de la Información. En este artículo se presentará el nat y el ban (también deno- minado hartley), y se relacionarán con el bit. Objetivos y conocimientos previos Los objetivos de aprendizaje este artículo docente se presentan en la tabla 1. Aunque se trata de un artículo divulgativo, cualquier conocimiento sobre Teo- ría de la Información que tenga el lector le será útil para asimilar mejor el contenido. 1. Mostrar unidades alternativas al bit para medir la cantidad de información y la entropía 2. Presentar las unidades ban (o hartley) y nat 3. Contextualizar cada unidad con la historia de la Teoría de la Información 4. Utilizar la Wikipedia como herramienta de referencia 5. Animar a la edición de la Wikipedia, completando los artículos en español, una vez aprendidos los conceptos presentados y su contexto histórico Tabla 1: Objetivos de aprendizaje del artículo docente 1 Introducción: bits El bit es utilizado de forma universal como unidad de medida básica de canti- dad de información. Su nombre proviene de la composición de las dos palabras binary digit (dígito binario, en inglés) por ser históricamente utilizado para denominar cada uno de los dos posibles valores binarios utilizados en compu- tación (0 y 1).
    [Show full text]
  • Shannon's Formula Wlog(1+SNR): a Historical Perspective
    Shannon’s Formula W log(1 + SNR): A Historical· Perspective on the occasion of Shannon’s Centenary Oct. 26th, 2016 Olivier Rioul <[email protected]> Outline Who is Claude Shannon? Shannon’s Seminal Paper Shannon’s Main Contributions Shannon’s Capacity Formula A Hartley’s rule C0 = log2 1 + ∆ is not Hartley’s 1 P Many authors independently derived C = 2 log2 1 + N in 1948. Hartley’s rule is exact: C0 = C (a coincidence?) C0 is the capacity of the “uniform” channel Shannon’s Conclusion 2 / 85 26/10/2016 Olivier Rioul Shannon’s Formula W log(1 + SNR): A Historical Perspective April 30, 2016 centennial day celebrated by Google: here Shannon is juggling with bits (1,0,0) in his communication scheme “father of the information age’’ Claude Shannon (1916–2001) 100th birthday 2016 April 30, 1916 Claude Elwood Shannon was born in Petoskey, Michigan, USA 3 / 85 26/10/2016 Olivier Rioul Shannon’s Formula W log(1 + SNR): A Historical Perspective here Shannon is juggling with bits (1,0,0) in his communication scheme “father of the information age’’ Claude Shannon (1916–2001) 100th birthday 2016 April 30, 1916 Claude Elwood Shannon was born in Petoskey, Michigan, USA April 30, 2016 centennial day celebrated by Google: 3 / 85 26/10/2016 Olivier Rioul Shannon’s Formula W log(1 + SNR): A Historical Perspective Claude Shannon (1916–2001) 100th birthday 2016 April 30, 1916 Claude Elwood Shannon was born in Petoskey, Michigan, USA April 30, 2016 centennial day celebrated by Google: here Shannon is juggling with bits (1,0,0) in his communication scheme “father of the information age’’ 3 / 85 26/10/2016 Olivier Rioul Shannon’s Formula W log(1 + SNR): A Historical Perspective “the most important man..
    [Show full text]
  • Quantum Computing and a Unified Approach to Fast Unitary Transforms
    Quantum Computing and a Unified Approach to Fast Unitary Transforms Sos S. Agaiana and Andreas Klappeneckerb aThe City University of New York and University of Texas at San Antonio [email protected] bTexas A&M University, Department of Computer Science, College Station, TX 77843-3112 [email protected] ABSTRACT A quantum computer directly manipulates information stored in the state of quantum mechanical systems. The available operations have many attractive features but also underly severe restrictions, which complicate the design of quantum algorithms. We present a divide-and-conquer approach to the design of various quantum algorithms. The class of algorithm includes many transforms which are well-known in classical signal processing applications. We show how fast quantum algorithms can be derived for the discrete Fourier transform, the Walsh-Hadamard transform, the Slant transform, and the Hartley transform. All these algorithms use at most O(log2 N) operations to transform a state vector of a quantum computer of length N. 1. INTRODUCTION Discrete orthogonal transforms and discrete unitary transforms have found various applications in signal, image, and video processing, in pattern recognition, in biocomputing, and in numerous other areas [1–11]. Well- known examples of such transforms include the discrete Fourier transform, the Walsh-Hadamard transform, the trigonometric transforms such as the Sine and Cosine transform, the Hartley transform, and the Slant transform. All these different transforms find applications in signal and image processing, because the great variety of signal classes occuring in practice cannot be handeled by a single transform. On a classical computer, the straightforward way to compute a discrete orthogonal transform of a signal vector of length N takes in general O(N 2) operations.
    [Show full text]
  • Revision of Lecture 5
    ELEC3203 Digital Coding and Transmission – Overview & Information Theory S Chen Revision of Lecture 5 Information transferring across channels • – Channel characteristics and binary symmetric channel – Average mutual information Average mutual information tells us what happens to information transmitted across • channel, or it “characterises” channel – But average mutual information is a bit too mathematical (too abstract) – As an engineer, one would rather characterises channel by its physical quantities, such as bandwidth, signal power and noise power or SNR Also intuitively given source with information rate R, one would like to know if • channel is capable of “carrying” the amount of information transferred across it – In other word, what is the channel capacity? – This lecture answers this fundamental question 71 ELEC3203 Digital Coding and Transmission – Overview & Information Theory S Chen A Closer Look at Channel Schematic of communication system • MODEM part X(k) Y(k) source sink x(t) y(t) {X i } {Y i } Analogue channel continuous−time discrete−time channel – Depending which part of system: discrete-time and continuous-time channels – We will start with discrete-time channel then continuous-time channel Source has information rate, we would like to know “capacity” of channel • – Moreover, we would like to know under what condition, we can achieve error-free transferring information across channel, i.e. no information loss – According to Shannon: information rate capacity error-free transfer ≤ → 72 ELEC3203 Digital Coding and Transmission – Overview & Information Theory S Chen Review of Channel Assumptions No amplitude or phase distortion by the channel, and the only disturbance is due • to additive white Gaussian noise (AWGN), i.e.
    [Show full text]
  • Nyquist, Shannon and the Information Carrying Capacity of Sig- Nals
    Nyquist, Shannon and the information carrying capacity of sig- nals Figure 1: The information highway There is whole science called the information theory. As far as a communications engineer is concerned, information is defined as a quantity called a bit. This is a pretty easy concept to intuit. A yes or a no, in or out, up or down, a 0 or a 1, these are all a form of information bits. In communications, a bit is a unit of information that can be conveyed by some change in the signal, be it amplitude, phase or frequency. We might issue a pulse of a fixed amplitude. The pulse of a positive amplitude can represent a 0 bit and one of a negative amplitude, a 1 bit. Let’s think about the transmission of information along the concepts of a highway as shown in Fig. 1. The width of this highway which is really the full electromagnetic spectrum, is infinite. The useful part of this highway is divided in lanes by regulatory bodies. We have very many lanes of different widths, which separate the multitude of users, such as satellites, microwave towers, wifi etc. In communications, we need to confine these signals in lanes. The specific lanes (i.e. of a certain center frequency) and their widths are assigned by ITU internationally and represent the allocated bandwidth. Within this width, the users can do what they want. In our highway analogy, we want our transporting truck to be the width of the lane, or in other words, we want the signal to fill the whole allocated bandwidth.
    [Show full text]
  • Maximal Repetition and Zero Entropy Rate
    Maximal Repetition and Zero Entropy Rate Łukasz Dębowski∗ Abstract Maximal repetition of a string is the maximal length of a repeated substring. This paper investigates maximal repetition of strings drawn from stochastic processes. Strengthening previous results, two new bounds for the almost sure growth rate of maximal repetition are identified: an upper bound in terms of conditional R´enyi entropy of order γ > 1 given a sufficiently long past and a lower bound in terms of unconditional Shannon entropy (γ = 1). Both the upper and the lower bound can be proved using an inequality for the distribution of recurrence time. We also supply an alternative proof of the lower bound which makes use of an inequality for the expectation of subword complexity. In particular, it is shown that a power-law logarithmic growth of maximal repetition with respect to the string length, recently observed for texts in natural language, may hold only if the conditional R´enyi entropy rate given a sufficiently long past equals zero. According to this observation, natural language cannot be faithfully modeled by a typical hidden Markov process, which is a class of basic language models used in computational linguistics. Keywords: maximal repetition, R´enyi entropies, entropy rate, recurrence time, subword complexity, natural language arXiv:1609.04683v4 [cs.IT] 26 Jul 2017 ∗Ł. Dębowski is with the Institute of Computer Science, Polish Academy of Sciences, ul. Jana Kazimierza 5, 01-248 Warszawa, Poland (e-mail: [email protected]). 1 I Motivation and main results n n Maximal repetition L(x1 ) of a string x1 = (x1; x2; :::; xn) is the maximal length of a repeated substring.
    [Show full text]
  • Photonic Qubits for Quantum Communication
    Photonic Qubits for Quantum Communication Exploiting photon-pair correlations; from theory to applications Maria Tengner Doctoral Thesis KTH School of Information and Communication Technology Stockholm, Sweden 2008 TRITA-ICT/MAP AVH Report 2008:13 KTH School of Information and ISSN 1653-7610 Communication Technology ISRN KTH/ICT-MAP/AVH-2008:13-SE Electrum 229 ISBN 978-91-7415-005-6 SE-164 40 Kista Sweden Akademisk avhandling som med tillstånd av Kungl Tekniska högskolan framlägges till offentlig granskning för avläggande av teknologie doktorsexamen i fotonik fredagen den 13 juni 2008 klockan 10.00 i Sal D, Forum, KTH Kista, Kungl Tekniska högskolan, Isafjordsgatan 39, Kista. © Maria Tengner, May 2008 iii Abstract For any communication, the conveyed information must be carried by some phys- ical system. If this system is a quantum system rather than a classical one, its behavior will be governed by the laws of quantum mechanics. Hence, the proper- ties of quantum mechanics, such as superpositions and entanglement, are accessible, opening up new possibilities for transferring information. The exploration of these possibilities constitutes the field of quantum communication. The key ingredient in quantum communication is the qubit, a bit that can be in any superposition of 0 and 1, and that is carried by a quantum state. One possible physical realization of these quantum states is to use single photons. Hence, to explore the possibilities of optical quantum communication, photonic quantum states must be generated, transmitted, characterized, and detected with high precision. This thesis begins with the first of these steps: the implementation of single-photon sources generat- ing photonic qubits.
    [Show full text]
  • Multiplicative Functions of Numbers Set and Logarithmic Identities
    Trends Journal of Sciences Research (2015) 2(1):13-16 http://www.tjsr.org Research Paper Open Access Multiplicative Functions of Numbers Set and Logarithmic Identities. Shannon and factorial logarithmic Identities, Entropy and Coentropy Yu. A. Kokotov Agrophysical Institute, Sankt- Petersburg, 195220, Russia *Correspondence: Yu. A. Kokotov([email protected]) Abstract The multiplicative functions characterizing the finite set of positive numbers are introduced in the work. With their help we find the logarithmic identities which connect logarithm of sum of the set numbers and logarithms of numbers themselves. One of them (contained in the work of Shannon) interconnects three information functions: information Hartley, entropy and coentropy. Shannon's identity allows better to understand the meaning and relationship of these collective characteristics of information (as the characteristics of finite sets and as probabilistic characteristics). The factorial multiplicative function and the logarithmic factorial identity are formed also from initial set numbers. That identity connects logarithms of factorials of integer numbers and logarithm of factorial of their sum. Keywords: Mumber Sets, Multiplicative Functions, Logarithmic Identity, The Identity of Shannon, Information Entropy, Coentropy, Factorials Yu. A. Kokotov. Multiplicative Functions of Numbers Set and Logarithmic Identities. Citation: Shannon and factorial logarithmic Identities, Entropy and Coentropy. Vol. 2, No. 1, 2015, pp.13-16 1. Introduction In information theory the logarithm of the sum, M, and the logarithms of its terms are considered as a The logarithmic scale is widely used in many areas of measure of the information. These ones are respectively science, including information theory. In this theory the measure of information about the set as a whole positive numbers logarithms are treated as Hartley (Hartley information, I0), and of information about each information about a set divided into subsets.The widely of the subsets (partial information, Ii).
    [Show full text]
  • The Case for Shifting the Renyi Entropy
    Article The case for shifting the Renyi Entropy Francisco J. Valverde-Albacete1,‡ ID , Carmen Peláez-Moreno2,‡* ID 1 Department of Signal Theory and Communications, Universidad Carlos III de Madrid, Leganés 28911, Spain; [email protected] 2 Department of Signal Theory and Communications, Universidad Carlos III de Madrid, Leganés 28911, Spain; [email protected] * Correspondence: [email protected]; Tel.: +34-91-624-8771 † These authors contributed equally to this work. Academic Editor: name Received: date; Accepted: date; Published: date Abstract: We introduce a variant of the Rényi entropy definition that aligns it with the well-known Hölder mean: in the new formulation, the r-th order Rényi Entropy is the logarithm of the inverse of the r-th order Hölder mean. This brings about new insights into the relationship of the Rényi entropy to quantities close to it, like the information potential and the partition function of statistical mechanics. We also provide expressions that allow us to calculate the Rényi entropies from the Shannon cross-entropy and the escort probabilities. Finally, we discuss why shifting the Rényi entropy is fruitful in some applications. Keywords: shifted Rényi entropy; Shannon-type relations; generalized weighted means; Hölder means; escort distributions. 1. Introduction Let X ∼ PX be a random variable over a set of outcomes X = fxi j 1 ≤ i ≤ ng and pmf PX defined in terms of the non-null values pi = PX(xi) . The Rényi entropy for X is defined in terms of that of PX as Ha(X) = Ha(PX) by a case analysis [1] 1 n a 6= 1 H (P ) = log pa (1) a X − ∑ i 1 a i=1 a = 1 lim Ha(PX) = H(PX) a!1 n where H(PX) = − ∑i=1 pi log pi is the Shannon entropy [2–4].
    [Show full text]
  • On Shannon and “Shannon's Formula”
    On Shannon and “Shannon’s formula” Lars Lundheim Department of Telecommunication, Norwegian University of Science and Technology (NTNU) The period between the middle of the nineteenth and the middle of the twentieth century represents a remarkable period in the history of science and technology. During this epoch, several discoveries and inventions removed many practical limitations of what individuals and societies could achieve. Especially in the field of communications, revolutionary developments took place such as high speed railroads, steam ships, aviation and telecommunications. It is interesting to note that as practical limitations were removed, several fundamental or principal limitations were established. For instance, Carnot showed that there was a fundamental limit to how much energy could be extracted from a heat engine. Later this result was generalized to the second law of thermodynamics. As a result of Einstein’s special relativity theory, the existence of an upper velocity limit was found. Other examples include Kelvin’s absolute zero, Heissenberg’s uncertainty principle and Gödel’s incompleteness theorem in mathematics. Shannon’s Channel coding theorem, which was published in 1948, seems to be the last one of such fundamental limits, and one may wonder why all of them were discovered during this limited time-span. One reason may have to do with maturity. When a field is young, researchers are eager to find out what can be done – not to identify borders they cannot pass. Since telecommunications is one of the youngest of the applied sciences, it is natural that the more fundamental laws were established at a late stage. In the present paper we will try to shed some light on developments that led up to Shannon’s information theory.
    [Show full text]