An Optimizing Up-and-Down Design

A central question in information theory is to determine the maximum success probability that can be achieved in sending a fixed number of messages over a noisy channel. This was first studied in the pioneering work of Shannon who established a simple expression characterizing this quantity in the limit of multiple independent uses of the channel. Here we consider the general setting with only one use of the channel. We observe that the maximum success probability can be expressed as the maximum value of a submodular function. Using this connection, we establish the following results: 1. There is a simple greedy polynomial-time algorithm that computes a code achieving a (1-1/e)-approximation of the maximum success probability. Moreover, for this problem it is NP-hard to obtain an approximation ratio strictly better than (1-1/e). 2. Shared quantum entanglement between the sender and the receiver can increase the success probability by a factor of at most 1/(1-1/e). In addition, this factor is tight if one allows an arbitrary non-signaling box between the sender and the receiver. 3. We give tight bounds on the one-shot performance of the meta-converse of Polyanskiy-Poor-Verdu.

The main contributions of this paper are two-fold.Firstly, we present a novel direction in the cryptanalysis of the cryptographic hash function SHA-1. Our work builds on previous cryptanalytic efforts on SHA-1 based on combinations of local collisions. Due to dependencies, previous approaches used heuristic corrections when combining the success probabilities and message conditions of the individual local collisions. Although this leads to success probabilities that are seemingly sufficient for feasible collision attacks, this approach most often does not lead to the maximum success probability possible as desired. We introduce novel techniques that enable us to determine the theoretical maximum success probability for a given set of (dependent) local collisions, as well as the smallest set of message conditions that attains this probability. We apply our new techniques and present an implemented open-source near-collision attack on SHA-1 with a complexity equivalent to 257.5 SHA-1 compressions.Secondly, we present an identical-prefix collision attack and a chosen-prefix collision attack on SHA-1 with complexities equivalent to approximately 261 and 277.1 SHA-1 compressions, respectively.

We study asymptotic behavior of stochastic approximation procedures with three main characteristics: truncations with random moving bounds, a matrix-valued random step-size sequence, and a dynamically changing random regression function. In particular, we show that under quitemild conditions, stochastic approximation procedures are asymptotically linear in the statistical sense, that is, they can be represented as weighted sums of random variables. Therefore, a suitable formof the central limit theoremcan be applied to derive asymptotic distribution of the corresponding processes. The theory is illustrated by various examples and special cases.

Channel simulation using coherence, which refers to realizing a target channel with coherent states and free operations, is a fundamental problem in the quantum resource theory of coherence. The limitations of the accuracy of deterministic channel simulation motivate us to consider the more general probabilistic framework. In this paper, we develop the framework for probabilistic channel simulation using coherence with free operations. When the chosen set of free operations is the maximally incoherent operations, we provide an efficiently computable semidefinite program (SDP) to calculate the maximal success probability and derive the analytic expression of success probability for some special cases. When the chosen set of free operations is the dephasing-covariant incoherent operations (DIO), we show that if the target channel is not a resource nonactivating channel then one cannot simulate it exactly both deterministically and probabilistically. The SDP for maximal success probability of simulating a channel by DIO is also given correspondingly. Published by the American Physical Society 2024

This paper is concerned with the strong convergence of recursive estimators which are generalizations of the Robbins-Monro (1951) stochastic approximation procedure. The Robbins-Monro procedure and its generalizations have been investigated in mаnу contexts, for example recursive nonlinear regression (Albert and Gardner, 1967), recursive maximum likelihood estimation (Fabian, 1978), robust estimation of parameters for autoregressive process (Campbell, 1982), robust estimation of a location parameter (Martin and Masreliez, 1975 and Holst, 1980, 1982), control of physical processes (Comer, 1964 and Ruppert, 1979, 1981), and system identification (Kushner and Clark, 1978, section 2.6). In this paper, we present a rather general convergence theorem which allows the disturbances to be dependent and enter in a non-additive fashion. As examples, the theorem is applied to the nonlinear regression estimator of Albert and Gardner (1967), and to the author’s (Ruppert 1979, 1981) Robbins-Monro type procedures for use where the root of the unknown regression function varies with time.KeywordsStochastic ApproximationAutoregressive ProcessStochastic Approximation AlgorithmStochastic Approximation MethodRandom Probability MeasureThese keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.

By using the theory of cavity QED, we study the system in which a two-level atom interacts with a cavity in the case of large detuning. Through the selective detecting of atomic state, Schrödinger cat states and entangled coherent states are easily generated. When the atom is driven by a weak classical field and the cavity field is in the Schrödinger cat state, we study the conditions of generating the Fock states and the maximal success probability. The maximal success probability in our scheme is larger than the previous one.

Linear multidimensional stochastic approximation procedures in continuous time with martingale errors are considered. An asymptotic behavior of the estimator obtained by trajectory averaging is studied. An asymptotics of integrated squared deviations functionals of the averaged estimator is found. Some results concerning fixed-size confidence regions are presented.

Two stochastic approximation procedures are proposed for finding a point attaining the maximum of a regression function defined and observable only at points on a set of discrete variables. The asymptotic convergence property of the procedures is discussed using the theorem of almost supermartingales. The procedures are applied to the recursive identification of autoregressive time series models. The identification procedure consists of a recursive order estimation stage and a recursive autoregressive parameter updating stage, and gives the true autoregressive model or the best autoregressive approximation model.

The ability to distill quantum coherence is pivotal for optimizing the performance of quantum technologies; however, such a task cannot always be accomplished with certainty. Here we develop a general framework of probabilistic distillation of quantum coherence in a one-shot setting, establishing fundamental limitations for different classes of free operations. We first provide a geometric interpretation for the maximal success probability, showing that under maximally incoherent operations (MIO) and dephasing-covariant incoherent operations (DIO) the problem can be simplified into efficiently computable semidefinite programs. Exploiting these results, we find that DIO and its subset of strictly incoherent operations have equal power in the probabilistic distillation of coherence from pure input states, while MIO are strictly stronger. We then prove a fundamental no-go result: Distilling coherence from any full-rank state is impossible even probabilistically. We further find that in some conditions the maximal success probability can vanish suddenly beyond a certain threshold in the distillation fidelity. Finally, we consider probabilistic coherence distillation assisted by a catalyst and demonstrate, with specific examples, its superiority to the unassisted and deterministic cases.

Motivated by the recent success of reinforcement learning algorithms, this paper studies a class of biased stochastic approximation (SA) procedures under a mild "ergodicity-like" assumption on the random noise sequence. Building on a multistep Lyapunov function that looks ahead to several future updates to accommodate the stochastic perturbations (thus gaining control over the bias), we prove a general result on the convergence of the SA iterates, and use it to derive non-asymptotic bounds on the mean-square error in the case of constant stepsizes. This novel viewpoint renders finite-time analysis of biased SA algorithms under a family of stochastic perturbations possible. For direct comparison with prior work, we demonstrate these bounds by applying them to TD-learning with linear function approximation, under the Markov chain observation model. The resultant finite-time error bound for TD-learning is the first of its kind, in the sense that it holds i) for the unmodified versions (i.e., without any modification to the updates) using even nonlinear approximators; as well as for Markov chains ii) under sublinear mixing conditions and iii) starting from any initial distribution, at least one of which has to be violated for existing results to be applicable.

This work develops a class of stochastic optimization algorithms for pricing American put options. The stock model is a regime-switching geometric Brownian motion. The switching process represents macro market states such as market trends, interest rates, etc. The solutions of pricing American options may be characterized by certain threshold values. Here, we show how one can use a stochastic approximation (SA) method to determine the optimal threshold levels. For option pricing in a finite horizon, a SA procedure is carried for a fixed time T. As T varies, the optimal threshold values obtained using stochastic approximation trace out a curve, called the threshold frontier. Convergence and rates of convergence are obtained using weak convergence methods and martingale averaging techniques. The proposed approach provides us with a viable computational approach, and has advantage in terms of the reduced computational complexity compared with the variational or quasi-variational inequality approach for optimal stopping.

CONVERGENCE PROPERTIES FOR A FAMILY OF BOUNDED FIXED STEP-SIZE ALGORITHMS

In the work presented here, we consider state and parameter estimation of a semi-nonlinear Markov jump system in a non-Gaussian noise environment. The non-Gaussian measurement noise is approximated by a finite Gaussian mixture model (GMM). We propose a maximum likelihood (ML) solution to this state estimation problem which leads to two expectation-maximization (EM) algorithms. The first is a batch EM method which takes all the available data in the conditional expectation of the state in the E-step. An interacting multiple model (IMM) smoother is employed to evaluate the conditional expectation of the state by which a suboptimal estimate of system state is directly obtained. The Gaussian mixture parameters are then updated in the M-step. The second is a recursive EM algorithm which results from a stochastic approximation procedure and uses a standard IMM filter. For performance evaluation, posterior Cramer-Rao bound (PCRB) on the state estimation is adopted. Simulation results verify the effectiveness of the proposed algorithms.

In this work the recursive EM algorithm is studied under the framework of array processing. The recursive EM algorithm is a stochastic approximation procedure with a special gain matrix. Under proper conditions, this procedure leads to strong consistency and asymptotic normality. Using a stochastic signal model, we have derived the recursive EM algorithm to find ML DOA estimates. In contrast to existing methods of finding ML estimates, the proposed method provides a quick update of the estimate as a new data enters. In addition, it can be applied to both narrow band and wideband signals. Numerical results show that the proposed algorithm performs well for closely located sources and low SNRs. This demonstrates the potential of the suggested algorithm in real time processing.

Along with the popular use of algorithms such as persistent contrastive divergence, tempered transition and parallel tempering, the past decade has witnessed a revival of learning undirected graphical models (UGMs) with sampling-based approximations. In this paper, based upon the analogy between Robbins-Monro's stochastic approximation procedure and sequential Monte Carlo (SMC), we analyze the strengths and limitations of state-of-the-art learning algorithms from an SMC point of view. Moreover, we apply the rationale further in sampling at each iteration, and propose to learn UGMs using persistent sequential Monte Carlo (PSMC). The whole learning procedure is based on the samples from a long, persistent sequence of distributions which are actively constructed. Compared to the above-mentioned algorithms, one critical strength of PSMC-based learning is that it can explore the sampling space more effectively. In particular, it is robust when learning rates are large or model distributions are high-dimensional and thus multi-modal, which often causes other algorithms to deteriorate. We tested PSMC learning, comparing it with related methods, on carefully designed experiments with both synthetic and real-world data. Our empirical results demonstrate that PSMC compares favorably with the state of the art by consistently yielding the highest (or among the highest) likelihoods. We also evaluated PSMC on two practical tasks, multi-label classification and image segmentation, in which PSMC displays promising applicability by outperforming others.