-
Have it your way: Individualized Privacy Assignment for DP-SGD
Authors:
Franziska Boenisch,
Christopher Mühl,
Adam Dziedzic,
Roy Rinberg,
Nicolas Papernot
Abstract:
When training a machine learning model with differential privacy, one sets a privacy budget. This budget represents a maximal privacy violation that any user is willing to face by contributing their data to the training set. We argue that this approach is limited because different users may have different privacy expectations. Thus, setting a uniform privacy budget across all points may be overly…
▽ More
When training a machine learning model with differential privacy, one sets a privacy budget. This budget represents a maximal privacy violation that any user is willing to face by contributing their data to the training set. We argue that this approach is limited because different users may have different privacy expectations. Thus, setting a uniform privacy budget across all points may be overly conservative for some users or, conversely, not sufficiently protective for others. In this paper, we capture these preferences through individualized privacy budgets. To demonstrate their practicality, we introduce a variant of Differentially Private Stochastic Gradient Descent (DP-SGD) which supports such individualized budgets. DP-SGD is the canonical approach to training models with differential privacy. We modify its data sampling and gradient noising mechanisms to arrive at our approach, which we call Individualized DP-SGD (IDP-SGD). Because IDP-SGD provides privacy guarantees tailored to the preferences of individual users and their data points, we find it empirically improves privacy-utility trade-offs.
△ Less
Submitted 19 January, 2024; v1 submitted 29 March, 2023;
originally announced March 2023.
-
Individualized PATE: Differentially Private Machine Learning with Individual Privacy Guarantees
Authors:
Franziska Boenisch,
Christopher Mühl,
Roy Rinberg,
Jannis Ihrig,
Adam Dziedzic
Abstract:
Applying machine learning (ML) to sensitive domains requires privacy protection of the underlying training data through formal privacy frameworks, such as differential privacy (DP). Yet, usually, the privacy of the training data comes at the cost of the resulting ML models' utility. One reason for this is that DP uses one uniform privacy budget epsilon for all training data points, which has to al…
▽ More
Applying machine learning (ML) to sensitive domains requires privacy protection of the underlying training data through formal privacy frameworks, such as differential privacy (DP). Yet, usually, the privacy of the training data comes at the cost of the resulting ML models' utility. One reason for this is that DP uses one uniform privacy budget epsilon for all training data points, which has to align with the strictest privacy requirement encountered among all data holders. In practice, different data holders have different privacy requirements and data points of data holders with lower requirements can contribute more information to the training process of the ML models. To account for this need, we propose two novel methods based on the Private Aggregation of Teacher Ensembles (PATE) framework to support the training of ML models with individualized privacy guarantees. We formally describe the methods, provide a theoretical analysis of their privacy bounds, and experimentally evaluate their effect on the final model's utility using the MNIST, SVHN, and Adult income datasets. Our empirical results show that the individualized privacy methods yield ML models of higher accuracy than the non-individualized baseline. Thereby, we improve the privacy-utility trade-off in scenarios in which different data holders consent to contribute their sensitive data at different individual privacy levels.
△ Less
Submitted 8 November, 2022; v1 submitted 21 February, 2022;
originally announced February 2022.
-
Poly-Sarcosine and Poly(ethylene-glycol) interactions with proteins investigated using molecular dynamics simulations
Authors:
Giovanni Settanni,
Timo Schäfer,
Christian Muhl,
Matthias Barz,
Friederike Schmid
Abstract:
Nanoparticles coated with hydrophilic polymers often show a reduction in unspecific interactions with the biological environment, which improves their biocompatibility. The molecular determinants of this reduction are not very well understood yet, and their knowledge may help improving nanoparticle design. Here we address, using molecular dynamics simulations, the interactions of human serum album…
▽ More
Nanoparticles coated with hydrophilic polymers often show a reduction in unspecific interactions with the biological environment, which improves their biocompatibility. The molecular determinants of this reduction are not very well understood yet, and their knowledge may help improving nanoparticle design. Here we address, using molecular dynamics simulations, the interactions of human serum albumin, the most abundant serum protein, with two promising hydrophilic polymers used for the coating of therapeutic nanoparticles, poly(ethylene-glycol) and poly-sarcosine. By simulating the protein immersed in a polymer-water mixture, we show that the two polymers have a very similar affinity for the protein surface, both in terms of the amount of polymer adsorbed and also in terms of the type of amino acids mainly involved in the interactions. We further analyze the kinetics of adsorption and how it affects the polymer conformations. Minor differences between the polymers are observed in the thickness of the adsorption layer, that are related to the different degree of flexibility of the two molecules. In comparison poly-alanine, an isomer of poly-sarcosine known to self-aggregate and induce protein aggregation, shows a significantly larger affinity for the protein surface than PEG and PSar, which we show to be related not to a different patterns of interactions with the protein surface, but to the different way the polymer interacts with water.
△ Less
Submitted 29 October, 2018;
originally announced October 2018.
-
Test Beam Performance Measurements for the Phase I Upgrade of the CMS Pixel Detector
Authors:
M. Dragicevic,
M. Friedl,
J. Hrubec,
H. Steininger,
A. Gädda,
J. Härkönen,
T. Lampén,
P. Luukka,
T. Peltola,
E. Tuominen,
E. Tuovinen,
A. Winkler,
P. Eerola,
T. Tuuva,
G. Baulieu,
G. Boudoul,
L. Caponetto,
C. Combaret,
D. Contardo,
T. Dupasquier,
G. Gallbit,
N. Lumb,
L. Mirabito,
S. Perries,
M. Vander Donckt
, et al. (462 additional authors not shown)
Abstract:
A new pixel detector for the CMS experiment was built in order to cope with the instantaneous luminosities anticipated for the Phase~I Upgrade of the LHC. The new CMS pixel detector provides four-hit tracking with a reduced material budget as well as new cooling and powering schemes. A new front-end readout chip mitigates buffering and bandwidth limitations, and allows operation at low comparator…
▽ More
A new pixel detector for the CMS experiment was built in order to cope with the instantaneous luminosities anticipated for the Phase~I Upgrade of the LHC. The new CMS pixel detector provides four-hit tracking with a reduced material budget as well as new cooling and powering schemes. A new front-end readout chip mitigates buffering and bandwidth limitations, and allows operation at low comparator thresholds. In this paper, comprehensive test beam studies are presented, which have been conducted to verify the design and to quantify the performance of the new detector assemblies in terms of tracking efficiency and spatial resolution. Under optimal conditions, the tracking efficiency is $99.95\pm0.05\,\%$, while the intrinsic spatial resolutions are $4.80\pm0.25\,μ\mathrm{m}$ and $7.99\pm0.21\,μ\mathrm{m}$ along the $100\,μ\mathrm{m}$ and $150\,μ\mathrm{m}$ pixel pitch, respectively. The findings are compared to a detailed Monte Carlo simulation of the pixel detector and good agreement is found.
△ Less
Submitted 1 June, 2017;
originally announced June 2017.
-
Performance of the EUDET-type beam telescopes
Authors:
H. Jansen,
S. Spannagel,
J. Behr,
A. Bulgheroni,
G. Claus,
E. Corrin,
D. G. Cussans,
J. Dreyling-Eschweiler,
D. Eckstein,
T. Eichhorn,
M. Goffe,
I. M. Gregor,
D. Haas,
C. Muhl,
H. Perrey,
R. Peschke,
P. Roloff,
I. Rubinskiy,
M. Winter
Abstract:
Test beam measurements at the test beam facilities of DESY have been conducted to characterise the performance of the EUDET-type beam telescopes originally developed within the EUDET project. The beam telescopes are equipped with six sensor planes using MIMOSA26 monolithic active pixel devices. A programmable Trigger Logic Unit provides trigger logic and time stamp information on particle passage.…
▽ More
Test beam measurements at the test beam facilities of DESY have been conducted to characterise the performance of the EUDET-type beam telescopes originally developed within the EUDET project. The beam telescopes are equipped with six sensor planes using MIMOSA26 monolithic active pixel devices. A programmable Trigger Logic Unit provides trigger logic and time stamp information on particle passage. Both data acquisition framework and offline reconstruction software packages are available. User devices are easily integrable into the data acquisition framework via predefined interfaces.
The biased residual distribution is studied as a function of the beam energy, plane spacing and sensor threshold. Its standard deviation at the two centre pixel planes using all six planes for tracking in a 6\,GeV electron/positron-beam is measured to be $(2.88\,\pm\,0.08)\,\upmu\meter$.Iterative track fits using the formalism of General Broken Lines are performed to estimate the intrinsic resolution of the individual pixel planes. The mean intrinsic resolution over the six sensors used is found to be $(3.24\,\pm\,0.09)\,\upmu\meter$.With a 5\,GeV electron/positron beam, the track resolution halfway between the two inner pixel planes using an equidistant plane spacing of 20\,mm is estimated to $(1.83\,\pm\,0.03)\,\upmu\meter$ assuming the measured intrinsic resolution. Towards lower beam energies the track resolution deteriorates due to increasing multiple scattering. Threshold studies show an optimal working point of the MIMOSA26 sensors at a sensor threshold of between five and six times their RMS noise. Measurements at different plane spacings are used to calibrate the amount of multiple scattering in the material traversed and allow for corrections to the predicted angular scattering for electron beams.
△ Less
Submitted 10 May, 2016; v1 submitted 31 March, 2016;
originally announced March 2016.
-
Trap** in irradiated p-on-n silicon sensors at fluences anticipated at the HL-LHC outer tracker
Authors:
W. Adam,
T. Bergauer,
M. Dragicevic,
M. Friedl,
R. Fruehwirth,
M. Hoch,
J. Hrubec,
M. Krammer,
W. Treberspurg,
W. Waltenberger,
S. Alderweireldt,
W. Beaumont,
X. Janssen,
S. Luyckx,
P. Van Mechelen,
N. Van Remortel,
A. Van Spilbeeck,
P. Barria,
C. Caillol,
B. Clerbaux,
G. De Lentdecker,
D. Dobur,
L. Favart,
A. Grebenyuk,
Th. Lenzi
, et al. (663 additional authors not shown)
Abstract:
The degradation of signal in silicon sensors is studied under conditions expected at the CERN High-Luminosity LHC. 200 $μ$m thick n-type silicon sensors are irradiated with protons of different energies to fluences of up to $3 \cdot 10^{15}$ neq/cm$^2$. Pulsed red laser light with a wavelength of 672 nm is used to generate electron-hole pairs in the sensors. The induced signals are used to determi…
▽ More
The degradation of signal in silicon sensors is studied under conditions expected at the CERN High-Luminosity LHC. 200 $μ$m thick n-type silicon sensors are irradiated with protons of different energies to fluences of up to $3 \cdot 10^{15}$ neq/cm$^2$. Pulsed red laser light with a wavelength of 672 nm is used to generate electron-hole pairs in the sensors. The induced signals are used to determine the charge collection efficiencies separately for electrons and holes drifting through the sensor. The effective trap** rates are extracted by comparing the results to simulation. The electric field is simulated using Synopsys device simulation assuming two effective defects. The generation and drift of charge carriers are simulated in an independent simulation based on PixelAV. The effective trap** rates are determined from the measured charge collection efficiencies and the simulated and measured time-resolved current pulses are compared. The effective trap** rates determined for both electrons and holes are about 50% smaller than those obtained using standard extrapolations of studies at low fluences and suggests an improved tracker performance over initial expectations.
△ Less
Submitted 7 May, 2015;
originally announced May 2015.
-
Review of the Use of Electroencephalography as an Evaluation Method for Human-Computer Interaction
Authors:
Jérémy Frey,
Christian Mühl,
Fabien Lotte,
Martin Hachet
Abstract:
Evaluating human-computer interaction is essential as a broadening population uses machines, sometimes in sensitive contexts. However, traditional evaluation methods may fail to combine real-time measures, an "objective" approach and data contextualization. In this review we look at how adding neuroimaging techniques can respond to such needs. We focus on electroencephalography (EEG), as it could…
▽ More
Evaluating human-computer interaction is essential as a broadening population uses machines, sometimes in sensitive contexts. However, traditional evaluation methods may fail to combine real-time measures, an "objective" approach and data contextualization. In this review we look at how adding neuroimaging techniques can respond to such needs. We focus on electroencephalography (EEG), as it could be handled effectively during a dedicated evaluation phase. We identify workload, attention, vigilance, fatigue, error recognition, emotions, engagement, flow and immersion as being recognizable by EEG. We find that workload, attention and emotions assessments would benefit the most from EEG. Moreover, we advocate to study further error recognition through neuroimaging to enhance usability and increase user experience.
△ Less
Submitted 9 November, 2013;
originally announced November 2013.
-
Infrastructure for Detector Research and Development towards the International Linear Collider
Authors:
J. Aguilar,
P. Ambalathankandy,
T. Fiutowski,
M. Idzik,
Sz. Kulis,
D. Przyborowski,
K. Swientek,
A. Bamberger,
M. Köhli,
M. Lupberger,
U. Renz,
M. Schumacher,
Andreas Zwerger,
A. Calderone,
D. G. Cussans,
H. F. Heath,
S. Mandry,
R. F. Page,
J. J. Velthuis,
D. Attié,
D. Calvet,
P. Colas,
X. Coppolani,
Y. Degerli,
E. Delagnes
, et al. (252 additional authors not shown)
Abstract:
The EUDET-project was launched to create an infrastructure for develo** and testing new and advanced detector technologies to be used at a future linear collider. The aim was to make possible experimentation and analysis of data for institutes, which otherwise could not be realized due to lack of resources. The infrastructure comprised an analysis and software network, and instrumentation infras…
▽ More
The EUDET-project was launched to create an infrastructure for develo** and testing new and advanced detector technologies to be used at a future linear collider. The aim was to make possible experimentation and analysis of data for institutes, which otherwise could not be realized due to lack of resources. The infrastructure comprised an analysis and software network, and instrumentation infrastructures for tracking detectors as well as for calorimetry.
△ Less
Submitted 23 January, 2012;
originally announced January 2012.
-
Design and Test of a Forward Neutron Calorimeter for the ZEUS Experiment
Authors:
The ZEUS FNC Group,
S. Bhadra,
I. Bohnet,
M. Cardy,
U. Dosselli,
C. -P. Fagerstroem,
W. Frisken,
K. Furutani,
D. Hanna,
U. Holm,
K. F. Johnson,
M. Khakzad,
G. Levman,
J. N. Lim,
B. Loehr,
J. F. Martin,
C. Muhl,
T. Neumann,
M. Rohde,
W. B. Schmidke,
D. G. Stairs,
H. Tiecke,
C. Voci
Abstract:
A lead scintillator sandwich sampling calorimeter has been installed in the HERA tunnel 105.6 m from the central ZEUS detector in the proton beam direction. It is designed to measure the energy and scattering angle of neutrons produced in charge exchange ep collisions. Before installation the calorimeter was tested and calibrated in the H6 beam at CERN where 120 GeV electrons, muons, pions and p…
▽ More
A lead scintillator sandwich sampling calorimeter has been installed in the HERA tunnel 105.6 m from the central ZEUS detector in the proton beam direction. It is designed to measure the energy and scattering angle of neutrons produced in charge exchange ep collisions. Before installation the calorimeter was tested and calibrated in the H6 beam at CERN where 120 GeV electrons, muons, pions and protons were made incident on the calorimeter. In addition, the spectrum of fast neutrons from charge exchange proton-lucite collisions was measured. The design and construction of the calorimeter is described, and the results of the CERN test reported. Special attention is paid to the measurement of shower position, shower width, and the separation of electromagnetic showers from hadronic showers. The overall energy scale as determined from the energy spectrum of charge exchange neutrons is compared to that obtained from direct beam hadrons.
△ Less
Submitted 27 January, 1997;
originally announced January 1997.
