From b.t.szabo at math.leidenuniv.nl Thu May 2 12:54:20 2019
From: b.t.szabo at math.leidenuniv.nl (Szabo, B.T.)
Date: Thu, 2 May 2019 10:54:20 +0000
Subject: [ml_ned] Reminder: Thematic seminar, Friday, 3pm-5pm, Leiden
Message-ID: <2B027EA14C9BFE4ABCC1C8257A16CB182E9D91AE@SPMXM05.VUW.leidenuniv.nl>
Dear Colleagues,
This is a kind reminder that the next Thematic Statistics Seminar meeting will be tomorrow (Friday, the 3rd of May) between 3 pm and 5 pm at Leiden University. This time we have two speakers: Madalin Guta (University of Nottingham) and Emilie Kaufmann (University of Lille) coming from abroad.
Speaker: Madalin Guta (University of Nottingham)
Title: An introduction to quantum state estimation
Time: 15:00-16:00, 3rd of May, 2019
Location: room 405, MI, Leiden University, 1 Niels Bohrweg, Leiden
Abstract:
Quantum state estimation is a key statistical tool for quantum engineering applications.
In quantum computation for instance, one applies a certain sequence of transformations (quantum algorithm) to a register of two dimensional quantum systems (qubits), which often produces joint states with special correlation properties (entanglement). In order to verify that the preparation procedure has been successful, one needs to measure the final state, to obtain a random outcome whose distribution depends on the quantum state. By repeating the procedure over many independent preparations, one collects statistical data that can be used to statistically reconstruct the state.
In this talk I will introduce the basic concepts required to formulation the state estimation problem (also known as quantum tomography), and present a recently proposed method for doing this. The Projected Least Squares (PLS) estimator consists of first computing a traditional least square estimator, which is subsequently projected onto the space of quantum states (positive trace-one matrices). This method is faster that traditional maximum likelihood estimation, and has good statistical behaviour. Indeed, using matrix concentration inequalities we show that PLS attains fundamental lower bounds for the estimation of low rank states with separate measurements.
Time permitting, I will discuss a more fundamental question concerning the highest estimation precision allowed by quantum mechanics. Here the concept of quantum local asymptotic normality (QLAN) gives a precise answer to this question and indicates what measurements are required to achieve this precision
-------------------
Speaker: Emilie Kaufmann (University of Lille)
Title: Generalized Likelihood Ratio Tests applied to Sequential Decision Making
Time: 16:00-17:00, 3rd of May, 2019
Location: room 405, MI, Leiden University, 1 Niels Bohrweg, Leiden
Abstract: In this talk I will discuss the use of a well-known statistical test, based the Generalized Likelihood Ratio (GLR), for different sequential decision making problems cast in the so-called multi-armed bandit framework. I a first part, I will focus on active testing in a bandit model, for which I will propose a choice of threshold for which a GLR-based stopping rule can be proved to return a correct answer with high probability. You will also understand why GLR are natural candidates for making a decision using a minimal number of samples. In a second part, I will show a different application of GLR tests to sequential decision making: GLR-based change-point detectors can be useful for minimizing regret in a piece-wise stationary multi-armed bandit.
Best wishes,
Tim van Erven, Moritz Schauer, and Botond Szabo
ps.: For more information please visit the seminar webpage: https://mschauer.github.io/StructuresSeminar/
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
From W.M.Koolen-Wijkstra at cwi.nl Wed May 29 18:19:19 2019
From: W.M.Koolen-Wijkstra at cwi.nl (Wouter Koolen-Wijkstra)
Date: Wed, 29 May 2019 18:19:19 +0200
Subject: [ml_ned] CWI Machine Learning seminar: Odalric Ambryn-Maillard Tue
4 Jun, 11:00, L120
Message-ID:
Dear all,
It is my pleasure to announce the following CWI Machine Learning seminar.
Speaker:? Odalric Ambryn-Maillard (INRIA Lille)
Title:??? A small tour of time-uniform concentration inequalities and
some open problems
Date:???? Tuesday 4 June, 11:00
Location: CWI L120
Please find the abstract below.
Hope to see you then.
Best wishes,
Wouter
Details:
https://portals.project.cwi.nl/ml-reading-group/events/tbd-odalric-ambryn-maillard
============
A small tour of time-uniform concentration inequalities and some open
problems
Odalric Ambryn-Maillard, INRIA Lille
In this talk, I will first discuss concentration inequalities that hold
uniformly over time, and more specifically different proof techniques
and some intriguing results. After illustrating this on a few examples,
I will then present some preliminary remarks about two questions. The
first one is that of adaptive variance estimation when facing model
misspecification. The second one is about aggregation of experts when
experts are learning agents using an underlying stochastic model.