-
Complex Clip** for Improved Generalization in Machine Learning
Authors:
Les Atlas,
Nicholas Rasmussen,
Felix Schwock,
Mert Pilanci
Abstract:
For many machine learning applications, a common input representation is a spectrogram. The underlying representation for a spectrogram is a short time Fourier transform (STFT) which gives complex values. The spectrogram uses the magnitude of these complex values, a commonly used detector. Modern machine learning systems are commonly overparameterized, where possible ill-conditioning problems are…
▽ More
For many machine learning applications, a common input representation is a spectrogram. The underlying representation for a spectrogram is a short time Fourier transform (STFT) which gives complex values. The spectrogram uses the magnitude of these complex values, a commonly used detector. Modern machine learning systems are commonly overparameterized, where possible ill-conditioning problems are ameliorated by regularization. The common use of rectified linear unit (ReLU) activation functions between layers of a deep net has been shown to help this regularization, improving system performance. We extend this idea of ReLU activation to detection for the complex STFT, providing a simple-to-compute modified and regularized spectrogram, which potentially results in better behaved training. We then confirmed the benefit of this approach on a noisy acoustic data set used for a real-world application. Generalization performance improved substantially. This approach might benefit other applications which use time-frequency map**s, for acoustic, audio, and other applications.
△ Less
Submitted 27 February, 2023;
originally announced February 2023.
-
Partitioning a Polygon Into Small Pieces
Authors:
Mikkel Abrahamsen,
Nichlas Langhoff Rasmussen
Abstract:
We study the problem of partitioning a given simple polygon $P$ into a minimum number of polygonal pieces, each of which has bounded size. We give algorithms for seven notions of `bounded size,' namely that each piece has bounded area, perimeter, straight-line diameter, geodesic diameter, or that each piece must be contained in a unit disk, an axis-aligned unit square or an arbitrarily rotated uni…
▽ More
We study the problem of partitioning a given simple polygon $P$ into a minimum number of polygonal pieces, each of which has bounded size. We give algorithms for seven notions of `bounded size,' namely that each piece has bounded area, perimeter, straight-line diameter, geodesic diameter, or that each piece must be contained in a unit disk, an axis-aligned unit square or an arbitrarily rotated unit square.
A more general version of the area problem has already been studied. Here we are, in addition to $P$, given positive real values $a_1,\ldots,a_k$ such that the sum $\sum_{i=1}^k a_i$ equals the area of $P$. The goal is to partition $P$ into exactly $k$ pieces $Q_1,\ldots,Q_k$ such that the area of $Q_i$ is $a_i$. Such a partition always exists, and an algorithm with running time $O(nk)$ has previously been described, where $n$ is the number of corners of $P$. We give an algorithm with optimal running time $O(n+k)$. For polygons with holes, we get running time $O(n\log n+k)$.
For the other problems, it seems out of reach to compute optimal partitions for simple polygons; for most of them, even in extremely restricted cases such as when $P$ is a square. We therefore develop $O(1)$-approximation algorithms for these problems, which means that the number of pieces in the produced partition is at most a constant factor larger than the cardinality of a minimum partition. Existing algorithms do not allow Steiner points, which means that all corners of the produced pieces must also be corners of $P$. This has the disappointing consequence that a partition does often not exist, whereas our algorithms always produce useful partitions. Furthermore, an optimal partition without Steiner points may require $Ω(n)$ pieces for polygons where a partition consisting of just $2$ pieces exists when Steiner points are allowed.
△ Less
Submitted 2 November, 2022;
originally announced November 2022.
-
Using Deep Learning with Large Aggregated Datasets for COVID-19 Classification from Cough
Authors:
Esin Darici Haritaoglu,
Nicholas Rasmussen,
Daniel C. H. Tan,
Jennifer Ranjani J.,
Jaclyn Xiao,
Gunvant Chaudhari,
Akanksha Rajput,
Praveen Govindan,
Christian Canham,
Wei Chen,
Minami Yamaura,
Laura Gomezjurado,
Aaron Broukhim,
Amil Khanzada,
Mert Pilanci
Abstract:
The Covid-19 pandemic has been one of the most devastating events in recent history, claiming the lives of more than 5 million people worldwide. Even with the worldwide distribution of vaccines, there is an apparent need for affordable, reliable, and accessible screening techniques to serve parts of the World that do not have access to Western medicine. Artificial Intelligence can provide a soluti…
▽ More
The Covid-19 pandemic has been one of the most devastating events in recent history, claiming the lives of more than 5 million people worldwide. Even with the worldwide distribution of vaccines, there is an apparent need for affordable, reliable, and accessible screening techniques to serve parts of the World that do not have access to Western medicine. Artificial Intelligence can provide a solution utilizing cough sounds as a primary screening mode for COVID-19 diagnosis. This paper presents multiple models that have achieved relatively respectable performance on the largest evaluation dataset currently presented in academic literature. Through investigation of a self-supervised learning model (Area under the ROC curve, AUC = 0.807) and a convolutional nerual network (CNN) model (AUC = 0.802), we observe the possibility of model bias with limited datasets. Moreover, we observe that performance increases with training data size, showing the need for the worldwide collection of data to help combat the Covid-19 pandemic with non-traditional means.
△ Less
Submitted 29 March, 2022; v1 submitted 5 January, 2022;
originally announced January 2022.
-
Improvement of Noise-Robust Single-Channel Voice Activity Detection with Spatial Pre-processing
Authors:
Max Væhrens,
Andreas Jonas Fuglsig,
Anders Post Jacobsen,
Nicolai Almskou Rasmussen,
Victor Mølbach Nissen,
Joachim Roland Hejslet,
Zheng-Hua Tan
Abstract:
Voice activity detection (VAD) remains a challenge in noisy environments. With access to multiple microphones, prior studies have attempted to improve the noise robustness of VAD by creating multi-channel VAD (MVAD) methods. However, MVAD is relatively new compared to single-channel VAD (SVAD), which has been thoroughly developed in the past. It might therefore be advantageous to improve SVAD meth…
▽ More
Voice activity detection (VAD) remains a challenge in noisy environments. With access to multiple microphones, prior studies have attempted to improve the noise robustness of VAD by creating multi-channel VAD (MVAD) methods. However, MVAD is relatively new compared to single-channel VAD (SVAD), which has been thoroughly developed in the past. It might therefore be advantageous to improve SVAD methods with pre-processing to obtain superior VAD, which is under-explored. This paper improves SVAD through two pre-processing methods, a beamformer and a spatial target speaker detector. The spatial detector sets signal frames to zero when no potential speaker is present within a target direction. The detector may be implemented as a filter, meaning the input signal for the SVAD is filtered according to the detector's output; or it may be implemented as a spatial VAD to be combined with the SVAD output. The evaluation is made on a noisy reverberant speech database, with clean speech from the Aurora 2 database and with white and babble noise. The results show that SVAD algorithms are significantly improved by the presented pre-processing methods, especially the spatial detector, across all signal-to-noise ratios. The SVAD algorithms with pre-processing significantly outperform a baseline MVAD in challenging noise conditions.
△ Less
Submitted 12 April, 2021;
originally announced April 2021.
-
Theoretical and experimental studies of photomechanical materials
Authors:
Bojun Zhou,
Elizabeth Bernhardt,
Ankita Bhuyan,
Zoya Ghorbanishiadeh,
Nathan Rasmussen,
Joseph Lanska,
Mark G. Kuzyk
Abstract:
After a brief introduction to the field of light-responsive materials, this paper provides a general theory for modeling the photomechanical response of a material, applies it to the two best-known mechanisms of photothermal heating and photo-isomerization, and then describes an experimental apparatus for quantitative measurements of the stress response. Several different materials are characteriz…
▽ More
After a brief introduction to the field of light-responsive materials, this paper provides a general theory for modeling the photomechanical response of a material, applies it to the two best-known mechanisms of photothermal heating and photo-isomerization, and then describes an experimental apparatus for quantitative measurements of the stress response. Several different materials are characterized to illustrate how the experiments and theory can be used to isolate the contributing mechanisms both through photomechanical measurements and auxiliary measurements of laser heating and thermal expansion. The efficiency and figure of merit of the photomechanical response is defined on several scales form the molecule to the bulk, and the photomorphon -- the basic material element that determines the bulk response -- is introduced. The photomorphon provides a conceptual model that can be expressed in terms of viscoelastic elements such as springs in series and parallel with the photoactive molecule. The photomechanical response, figure of merit, and the deduced microscopic photomechanical properties are tabulated and proposals for new materials classes are made.
△ Less
Submitted 1 February, 2019;
originally announced February 2019.
-
An apparatus for measuring a material's photomechanical response
Authors:
Elizabeth Bernhardt,
Chad Garrison,
Nathan Rasmussen,
Joseph Lanska,
Mark Kuzyk
Abstract:
Please find the final and published version of the manuscript using this citation: E. Bernhardt, et. al, "An apparatus for measuring a material's photomechanical response", American Journal of Physics, 86, 14 (2018)
This paper describes a simple and inexpensive apparatus for measuring the light-induced shape change of a material, which can be implemented in a high school or undergraduate laborat…
▽ More
Please find the final and published version of the manuscript using this citation: E. Bernhardt, et. al, "An apparatus for measuring a material's photomechanical response", American Journal of Physics, 86, 14 (2018)
This paper describes a simple and inexpensive apparatus for measuring the light-induced shape change of a material, which can be implemented in a high school or undergraduate laboratory. The key components are a laser pointer to actuate the material, a force sensor from an inexpensive jeweler's balance to measure the response, an Arduino for data acquisition and a means for mechanically mounting the components. The apparatus described here was used by high school students and teachers in a summer program to characterize liquid crystal elastomers. The theory of the photomechanical response is used to interpret the data, from which the material parameters are determined.
△ Less
Submitted 16 December, 2018; v1 submitted 25 June, 2018;
originally announced June 2018.
-
Leaderboard Effects on Player Performance in a Citizen Science Game
Authors:
Mads Kock Pedersen,
Nanna Ravn Rasmussen,
Jacob F. Sherson,
Rajiv Vaid Basaiawmoit
Abstract:
Quantum Moves is a citizen science game that investigates the ability of humans to solve complex physics challenges that are intractable for computers. During the launch of Quantum Moves in April 2016 the game's leaderboard function broke down resulting in a "no leaderboard" game experience for some players for a couple of days (though their scores were still displayed). The subsequent quick fix o…
▽ More
Quantum Moves is a citizen science game that investigates the ability of humans to solve complex physics challenges that are intractable for computers. During the launch of Quantum Moves in April 2016 the game's leaderboard function broke down resulting in a "no leaderboard" game experience for some players for a couple of days (though their scores were still displayed). The subsequent quick fix of an all-time Top 5 leaderboard, and the following long-term implementation of a personalized relative-position (infinite) leaderboard provided us with a unique opportunity to compare and investigate the effect of different leaderboard implementations on player performance in a points-driven citizen science game.
All three conditions were live sequentially during the game's initial influx of more than 150.000 players that stemmed from global press attention on Quantum Moves due the publication of a Nature paper about the use of Quantum Moves in solving a specific quantum physics problem. Thus, it has been possible to compare the three conditions and their influence on the performance (defined as a player's quality of game play related to a high-score) of over 4500 new players. These 4500 odd players in our three leaderboard-conditions have a similar demographic background based upon the time-window over which the implementations occurred and controlled against Player ID tags. Our results placed Condition 1 experience over condition 3 and in some cases even over condition 2 which goes against the general assumption that leaderboards enhance gameplay and its subsequent overuse as a an oft-relied upon element that designers slap onto a game to enhance said appeal. Our study thus questions the use of leaderboards as general performance enhancers in gamification contexts and brings some empirical rigor to an often under-reported but overused phenomenon.
△ Less
Submitted 12 July, 2017;
originally announced July 2017.
-
Pencils and nets on curves arising from rank 1 sheaves on K3 surfaces
Authors:
Nils Henry Rasmussen
Abstract:
Let $S$ be a K3 surface, $C$ a smooth curve on $S$ with $\mathcal{O}_S(C)$ ample, and $A$ a base-point free $g^2_d$ on $C$ of small degree. We use Lazarsfeld--Mukai bundles to prove that $A$ is cut out by the global sections of a rank 1 torsion-free sheaf $\mathcal{G}$ on $S$. Furthermore, we show that $c_1(\mathcal{G})$ with one exception is adapted to $\mathcal{O}_S(C)$ and satisfies…
▽ More
Let $S$ be a K3 surface, $C$ a smooth curve on $S$ with $\mathcal{O}_S(C)$ ample, and $A$ a base-point free $g^2_d$ on $C$ of small degree. We use Lazarsfeld--Mukai bundles to prove that $A$ is cut out by the global sections of a rank 1 torsion-free sheaf $\mathcal{G}$ on $S$. Furthermore, we show that $c_1(\mathcal{G})$ with one exception is adapted to $\mathcal{O}_S(C)$ and satisfies $\mathrm{Cliff}(c_1(\mathcal{G})_{|C})\leq\mathrm{Cliff}(A)$, thereby confirming a conjecture posed by Donagi and Morrison. We also show that the same methods can be used to give a simple proof of the conjecture in the $g^1_d$ case.
In the final section, we give an example of the mentioned exception where $h^0(C,c_1(\mathcal{G})_{|C})$ is dependent on the curve $C$ in its linear system, thereby failing to be adapted to $\mathcal{O}_S(C)$.
△ Less
Submitted 9 April, 2018; v1 submitted 5 November, 2015;
originally announced November 2015.
-
Pencils and nets of small degree on curves on smooth, projective surfaces of Picard rank 1 and very ample generator
Authors:
Nils Henry Rasmussen
Abstract:
Let S be a smooth, projective surface of Picard rank 1 and very ample generator embedding S into P^n. Let C be a smooth curve in O(m) for m \geq 5. We prove that any base-point free, complete g^r_d on C for r\in\{1,2\} and d small enough is cut out by a hyperplane section restricted to a multisecant (n-r-1)-plane.
Let S be a smooth, projective surface of Picard rank 1 and very ample generator embedding S into P^n. Let C be a smooth curve in O(m) for m \geq 5. We prove that any base-point free, complete g^r_d on C for r\in\{1,2\} and d small enough is cut out by a hyperplane section restricted to a multisecant (n-r-1)-plane.
△ Less
Submitted 18 August, 2015; v1 submitted 12 May, 2015;
originally announced May 2015.
-
Pencils of small degree on curves on unnodal Enriques surfaces
Authors:
Nils Henry Rasmussen,
Shengtian Zhou
Abstract:
We use vector-bundle techniques in order to compute $\dim W^1_d(C)$ where $C$ is general and smooth in a linear system on an unnodal Enriques surface. We furthermore find new examples of smooth curves on Enriques surfaces with an infinite number of $g^1_{\gon(C)}$'s.
We use vector-bundle techniques in order to compute $\dim W^1_d(C)$ where $C$ is general and smooth in a linear system on an unnodal Enriques surface. We furthermore find new examples of smooth curves on Enriques surfaces with an infinite number of $g^1_{\gon(C)}$'s.
△ Less
Submitted 27 August, 2013; v1 submitted 21 July, 2013;
originally announced July 2013.
