-
Nonparametric Variational Regularisation of Pretrained Transformers
Authors:
Fabio Fehr,
James Henderson
Abstract:
The current paradigm of large-scale pre-training and fine-tuning Transformer large language models has lead to significant improvements across the board in natural language processing. However, such large models are susceptible to overfitting to their training data, and as a result the models perform poorly when the domain changes. Also, due to the model's scale, the cost of fine-tuning the model…
▽ More
The current paradigm of large-scale pre-training and fine-tuning Transformer large language models has lead to significant improvements across the board in natural language processing. However, such large models are susceptible to overfitting to their training data, and as a result the models perform poorly when the domain changes. Also, due to the model's scale, the cost of fine-tuning the model to the new domain is large. Nonparametric Variational Information Bottleneck (NVIB) has been proposed as a regulariser for training cross-attention in Transformers, potentially addressing the overfitting problem. We extend the NVIB framework to replace all types of attention functions in Transformers, and show that existing pretrained Transformers can be reinterpreted as Nonparametric Variational (NV) models using a proposed identity initialisation. We then show that changing the initialisation introduces a novel, information-theoretic post-training regularisation in the attention mechanism, which improves out-of-domain generalisation without any training. This success supports the hypothesis that pretrained Transformers are implicitly NV Bayesian models.
△ Less
Submitted 1 December, 2023;
originally announced December 2023.
-
Learning to Abstract with Nonparametric Variational Information Bottleneck
Authors:
Melika Behjati,
Fabio Fehr,
James Henderson
Abstract:
Learned representations at the level of characters, sub-words, words and sentences, have each contributed to advances in understanding different NLP tasks and linguistic phenomena. However, learning textual embeddings is costly as they are tokenization specific and require different models to be trained for each level of abstraction. We introduce a novel language representation model which can lea…
▽ More
Learned representations at the level of characters, sub-words, words and sentences, have each contributed to advances in understanding different NLP tasks and linguistic phenomena. However, learning textual embeddings is costly as they are tokenization specific and require different models to be trained for each level of abstraction. We introduce a novel language representation model which can learn to compress to different levels of abstraction at different layers of the same model. We apply Nonparametric Variational Information Bottleneck (NVIB) to stacked Transformer self-attention layers in the encoder, which encourages an information-theoretic compression of the representations through the model. We find that the layers within the model correspond to increasing levels of abstraction and that their representations are more linguistically informed. Finally, we show that NVIB compression results in a model which is more robust to adversarial perturbations.
△ Less
Submitted 26 October, 2023;
originally announced October 2023.
-
A Variational AutoEncoder for Transformers with Nonparametric Variational Information Bottleneck
Authors:
James Henderson,
Fabio Fehr
Abstract:
We propose a VAE for Transformers by develo** a variational information bottleneck regulariser for Transformer embeddings. We formalise the embedding space of Transformer encoders as mixture probability distributions, and use Bayesian nonparametrics to derive a nonparametric variational information bottleneck (NVIB) for such attention-based embeddings. The variable number of mixture components s…
▽ More
We propose a VAE for Transformers by develo** a variational information bottleneck regulariser for Transformer embeddings. We formalise the embedding space of Transformer encoders as mixture probability distributions, and use Bayesian nonparametrics to derive a nonparametric variational information bottleneck (NVIB) for such attention-based embeddings. The variable number of mixture components supported by nonparametric methods captures the variable number of vectors supported by attention, and the exchangeability of our nonparametric distributions captures the permutation invariance of attention. This allows NVIB to regularise the number of vectors accessible with attention, as well as the amount of information in individual vectors. By regularising the cross-attention of a Transformer encoder-decoder with NVIB, we propose a nonparametric variational autoencoder (NVAE). Initial experiments on training a NVAE on natural language text show that the induced embedding space has the desired properties of a VAE for Transformers.
△ Less
Submitted 12 August, 2022; v1 submitted 27 July, 2022;
originally announced July 2022.
-
HyperMixer: An MLP-based Low Cost Alternative to Transformers
Authors:
Florian Mai,
Arnaud Pannatier,
Fabio Fehr,
Haolin Chen,
Francois Marelli,
Francois Fleuret,
James Henderson
Abstract:
Transformer-based architectures are the model of choice for natural language understanding, but they come at a significant cost, as they have quadratic complexity in the input length, require a lot of training data, and can be difficult to tune. In the pursuit of lower costs, we investigate simple MLP-based architectures. We find that existing architectures such as MLPMixer, which achieves token m…
▽ More
Transformer-based architectures are the model of choice for natural language understanding, but they come at a significant cost, as they have quadratic complexity in the input length, require a lot of training data, and can be difficult to tune. In the pursuit of lower costs, we investigate simple MLP-based architectures. We find that existing architectures such as MLPMixer, which achieves token mixing through a static MLP applied to each feature independently, are too detached from the inductive biases required for natural language understanding. In this paper, we propose a simple variant, HyperMixer, which forms the token mixing MLP dynamically using hypernetworks. Empirically, we demonstrate that our model performs better than alternative MLP-based models, and on par with Transformers. In contrast to Transformers, HyperMixer achieves these results at substantially lower costs in terms of processing time, training data, and hyperparameter tuning.
△ Less
Submitted 13 November, 2023; v1 submitted 7 March, 2022;
originally announced March 2022.
-
Validation of GEANT4 Monte Carlo Models with a Highly Granular Scintillator-Steel Hadron Calorimeter
Authors:
C. Adloff,
J. Blaha,
J. -J. Blaising,
C. Drancourt,
A. Espargilière,
R. Gaglione,
N. Geffroy,
Y. Karyotakis,
J. Prast,
G. Vouters,
K. Francis,
J. Repond,
J. Schlereth,
J. Smith,
L. Xia,
E. Baldolemar,
J. Li,
S. T. Park,
M. Sosebee,
A. P. White,
J. Yu,
T. Buanes,
G. Eigen,
Y. Mikami,
N. K. Watson
, et al. (148 additional authors not shown)
Abstract:
Calorimeters with a high granularity are a fundamental requirement of the Particle Flow paradigm. This paper focuses on the prototype of a hadron calorimeter with analog readout, consisting of thirty-eight scintillator layers alternating with steel absorber planes. The scintillator plates are finely segmented into tiles individually read out via Silicon Photomultipliers. The presented results are…
▽ More
Calorimeters with a high granularity are a fundamental requirement of the Particle Flow paradigm. This paper focuses on the prototype of a hadron calorimeter with analog readout, consisting of thirty-eight scintillator layers alternating with steel absorber planes. The scintillator plates are finely segmented into tiles individually read out via Silicon Photomultipliers. The presented results are based on data collected with pion beams in the energy range from 8GeV to 100GeV. The fine segmentation of the sensitive layers and the high sampling frequency allow for an excellent reconstruction of the spatial development of hadronic showers. A comparison between data and Monte Carlo simulations is presented, concerning both the longitudinal and lateral development of hadronic showers and the global response of the calorimeter. The performance of several GEANT4 physics lists with respect to these observables is evaluated.
△ Less
Submitted 15 June, 2014; v1 submitted 13 June, 2013;
originally announced June 2013.
-
Hadronic energy resolution of a highly granular scintillator-steel hadron calorimeter using software compensation techniques
Authors:
CALICE Collaboration,
C. Adloff,
J. Blaha,
J. -J. Blaising,
C. Drancourt,
A. Espargilière,
R. Gaglione,
N. Geffroy,
Y. Karyotakis,
J. Prast,
G. Vouters,
K. Francis,
J. Repond,
J. Smith,
L. Xia,
E. Baldolemar,
J. Li,
S. T. Park,
M. Sosebee,
A. P. White,
J. Yu,
T. Buanes,
G. Eigen,
Y. Mikami,
N. K. Watson
, et al. (142 additional authors not shown)
Abstract:
The energy resolution of a highly granular 1 m3 analogue scintillator-steel hadronic calorimeter is studied using charged pions with energies from 10 GeV to 80 GeV at the CERN SPS. The energy resolution for single hadrons is determined to be approximately 58%/sqrt(E/GeV}. This resolution is improved to approximately 45%/sqrt(E/GeV) with software compensation techniques. These techniques take advan…
▽ More
The energy resolution of a highly granular 1 m3 analogue scintillator-steel hadronic calorimeter is studied using charged pions with energies from 10 GeV to 80 GeV at the CERN SPS. The energy resolution for single hadrons is determined to be approximately 58%/sqrt(E/GeV}. This resolution is improved to approximately 45%/sqrt(E/GeV) with software compensation techniques. These techniques take advantage of the event-by-event information about the substructure of hadronic showers which is provided by the imaging capabilities of the calorimeter. The energy reconstruction is improved either with corrections based on the local energy density or by applying a single correction factor to the event energy sum derived from a global measure of the shower energy density. The application of the compensation algorithms to Geant4 simulations yield resolution improvements comparable to those observed for real data.
△ Less
Submitted 27 September, 2012; v1 submitted 17 July, 2012;
originally announced July 2012.
-
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.
-
Acoustic and optical variations during rapid downward motion episodes in the deep north-western Mediterranean Sea
Authors:
H. van Haren,
I. Taupier-Letage,
J. A. Aguilar,
A. Albert,
M. Anghinolfi,
G. Anton,
S. Anvar,
M. Ardid,
A. C. Assis Jesus,
T. Astraatmadja,
J. -J. Aubert,
R. Auer,
B. Baret,
S. Basa,
M. Bazzotti,
V. Bertin,
S. Biagi,
C. Bigongiari,
M. Bou-Cabof,
M. C. Bouwhuis,
A. Brown,
J. Brunner,
J. Busto,
F. Camarena,
A. Capone
, et al. (116 additional authors not shown)
Abstract:
An Acoustic Doppler Current Profiler (ADCP) was moored at the deep-sea site of the ANTARES neutrino telescope near Toulon, France, thus providing a unique opportunity to compare high-resolution acoustic and optical observations between 70 and 170 m above the sea bed at 2475 m. The ADCP measured downward vertical currents of magnitudes up to 0.03 m s-1 in late winter and early spring 2006. In the s…
▽ More
An Acoustic Doppler Current Profiler (ADCP) was moored at the deep-sea site of the ANTARES neutrino telescope near Toulon, France, thus providing a unique opportunity to compare high-resolution acoustic and optical observations between 70 and 170 m above the sea bed at 2475 m. The ADCP measured downward vertical currents of magnitudes up to 0.03 m s-1 in late winter and early spring 2006. In the same period, observations were made of enhanced levels of acoustic reflection, interpreted as suspended particles including zooplankton, by a factor of about 10 and of horizontal currents reaching 0.35 m s-1. These observations coincided with high light levels detected by the telescope, interpreted as increased bioluminescence. During winter 2006 deep dense-water formation occurred in the Ligurian subbasin, thus providing a possible explanation for these observations. However, the 10-20 days quasi-periodic episodes of high levels of acoustic reflection, light and large vertical currents continuing into the summer are not direct evidence of this process. It is hypothesized that the main process allowing for suspended material to be moved vertically later in the year is local advection, linked with topographic boundary current instabilities along the rim of the 'Northern Current'.
△ Less
Submitted 28 November, 2011;
originally announced November 2011.
-
A Fast Algorithm for Muon Track Reconstruction and its Application to the ANTARES Neutrino Telescope
Authors:
ANTARES collaboration,
J. A. Aguilar,
I. Al Samarai,
A. Albert,
M. Andre,
M. Anghinolfi,
G. Anton,
S. Anvar,
M. Ardid,
A. C. Assis Jesus,
T. Astraatmadja,
J-J. Aubert,
R. Auer,
B. Baret,
S. Basa,
M. Bazzotti,
V. Bertin,
S. Biagi,
C. Bigongiari,
C. Bogazzi,
M. Bou-Cabo,
M. C. Bouwhuis,
A. M. Brown,
J. Brunner,
J. Busto
, et al. (118 additional authors not shown)
Abstract:
An algorithm is presented, that provides a fast and robust reconstruction of neutrino induced upward-going muons and a discrimination of these events from downward-going atmospheric muon background in data collected by the ANTARES neutrino telescope. The algorithm consists of a hit merging and hit selection procedure followed by fitting steps for a track hypothesis and a point-like light source. I…
▽ More
An algorithm is presented, that provides a fast and robust reconstruction of neutrino induced upward-going muons and a discrimination of these events from downward-going atmospheric muon background in data collected by the ANTARES neutrino telescope. The algorithm consists of a hit merging and hit selection procedure followed by fitting steps for a track hypothesis and a point-like light source. It is particularly well-suited for real time applications such as online monitoring and fast triggering of optical follow-up observations for multi-messenger studies. The performance of the algorithm is evaluated with Monte Carlo simulations and various distributions are compared with that obtained in ANTARES data.
△ Less
Submitted 20 May, 2011;
originally announced May 2011.
-
Electromagnetic response of a highly granular hadronic calorimeter
Authors:
C. Adloff,
J. Blaha,
J. -J. Blaising,
C. Drancourt,
A. Espargilière,
R. Gaglione,
N. Geffroy,
Y. Karyotakis,
J. Prast,
G. Vouters,
K. Francis,
J. Repond,
J. Smith,
L. Xia,
E. Baldolemar,
J. Li,
S. T. Park,
M. Sosebee,
A. P. White,
J. Yu,
Y. Mikami,
N. K. Watson T. Goto,
G. Mavromanolakis,
M. A. Thomson,
D. R. Ward W. Yan
, et al. (142 additional authors not shown)
Abstract:
The CALICE collaboration is studying the design of high performance electromagnetic and hadronic calorimeters for future International Linear Collider detectors. For the hadronic calorimeter, one option is a highly granular sampling calorimeter with steel as absorber and scintillator layers as active material. High granularity is obtained by segmenting the scintillator into small tiles individuall…
▽ More
The CALICE collaboration is studying the design of high performance electromagnetic and hadronic calorimeters for future International Linear Collider detectors. For the hadronic calorimeter, one option is a highly granular sampling calorimeter with steel as absorber and scintillator layers as active material. High granularity is obtained by segmenting the scintillator into small tiles individually read out via silicon photo-multipliers (SiPM).
A prototype has been built, consisting of thirty-eight sensitive layers, segmented into about eight thousand channels. In 2007 the prototype was exposed to positrons and hadrons using the CERN SPS beam, covering a wide range of beam energies and incidence angles. The challenge of cell equalization and calibration of such a large number of channels is best validated using electromagnetic processes.
The response of the prototype steel-scintillator calorimeter, including linearity and uniformity, to electrons is investigated and described.
△ Less
Submitted 8 June, 2011; v1 submitted 20 December, 2010;
originally announced December 2010.
-
Time Calibration of the ANTARES Neutrino Telescope
Authors:
The ANTARES Collaboration,
J. A. Aguilar,
I. Al Samarai,
A. Albert,
M. André,
M. Anghinolfi,
G. Anton,
S. Anvar,
M. Ardid,
A. C. Assis Jesus,
T. Astraatmadja,
J. J. Aubert,
R. Auer,
B. Baret,
S. Basa,
M. Bazzotti,
V. Bertin,
S. Biagi,
C. Bigongiari,
M. Bou-Cabo,
M. C. Bouwhuis,
A. M. Brown,
J. Brunner,
J. Busto,
F. Camarena
, et al. (113 additional authors not shown)
Abstract:
The ANTARES deep-sea neutrino telescope comprises a three-dimensional array of photomultipliers to detect the Cherenkov light induced by upgoing relativistic charged particles originating from neutrino interactions in the vicinity of the detector. The large scattering length of light in the deep sea facilitates an angular resolution of a few tenths of a degree for neutrino energies exceeding 10 Te…
▽ More
The ANTARES deep-sea neutrino telescope comprises a three-dimensional array of photomultipliers to detect the Cherenkov light induced by upgoing relativistic charged particles originating from neutrino interactions in the vicinity of the detector. The large scattering length of light in the deep sea facilitates an angular resolution of a few tenths of a degree for neutrino energies exceeding 10 TeV. In order to achieve this optimal performance, the time calibration procedures should ensure a relative time calibration between the photomultipliers at the level of about 1ns. The methods developed to attain this level of precision are described.
△ Less
Submitted 10 December, 2010;
originally announced December 2010.
-
Search for a diffuse flux of high-energy $ν_μ$ with the ANTARES neutrino telescope
Authors:
ANTARES Collaboration,
J. A. Aguilar,
I. Al Samarai,
A. Albert,
M. André,
M. Anghinolfi,
G. Anton,
S. Anvar,
M. Ardid,
A. C. Assis Jesus,
T. Astraatmadja,
J-J. Aubert,
R. Auer,
B. Baret,
S. Basa,
M. Bazzotti,
V. Bertin,
S. Biagi,
C. Bigongiari,
C. Bogazzi,
M. Bou-Cabo,
M. C. Bouwhuis,
A. M. Brown,
J. Brunner,
J. Busto
, et al. (119 additional authors not shown)
Abstract:
A search for a diffuse flux of astrophysical muon neutrinos, using data collected by the ANTARES neutrino telescope is presented. A $(0.83\times 2π)$ sr sky was monitored for a total of 334 days of equivalent live time. The searched signal corresponds to an excess of events, produced by astrophysical sources, over the expected atmospheric neutrino background. The observed number of events is found…
▽ More
A search for a diffuse flux of astrophysical muon neutrinos, using data collected by the ANTARES neutrino telescope is presented. A $(0.83\times 2π)$ sr sky was monitored for a total of 334 days of equivalent live time. The searched signal corresponds to an excess of events, produced by astrophysical sources, over the expected atmospheric neutrino background. The observed number of events is found compatible with the background expectation. Assuming an $E^{-2}$ flux spectrum, a 90% c.l. upper limit on the diffuse $ν_μ$ flux of $E^2Φ_{90%} = 5.3 \times 10^{-8} \ \mathrm{GeV\ cm^{-2}\ s^{-1}\ sr^{-1}} $ in the energy range 20 TeV - 2.5 PeV is obtained. Other signal models with different energy spectra are also tested and some rejected.
△ Less
Submitted 16 November, 2010;
originally announced November 2010.
-
AMADEUS - The Acoustic Neutrino Detection Test System of the ANTARES Deep-Sea Neutrino Telescope
Authors:
ANTARES collaboration,
J. A. Aguilar,
I. Al Samarai,
A. Albert,
M. Anghinolfi,
G. Anton,
S. Anvar,
M. Ardid,
A. C. Assis Jesus,
T. Astraatmadja,
J. -J. Aubert,
R. Auer,
E. Barbarito,
B. Baret,
S. Basa,
M. Bazzotti,
V. Bertin,
S. Biagi,
C. Bigongiari,
M. Bou-Cabo,
M. C. Bouwhuis,
A. Brown,
J. Brunner,
J. Busto,
F. Camarena
, et al. (127 additional authors not shown)
Abstract:
The AMADEUS (ANTARES Modules for the Acoustic Detection Under the Sea) system which is described in this article aims at the investigation of techniques for acoustic detection of neutrinos in the deep sea. It is integrated into the ANTARES neutrino telescope in the Mediterranean Sea. Its acoustic sensors, installed at water depths between 2050 and 2300 m, employ piezo-electric elements for the bro…
▽ More
The AMADEUS (ANTARES Modules for the Acoustic Detection Under the Sea) system which is described in this article aims at the investigation of techniques for acoustic detection of neutrinos in the deep sea. It is integrated into the ANTARES neutrino telescope in the Mediterranean Sea. Its acoustic sensors, installed at water depths between 2050 and 2300 m, employ piezo-electric elements for the broad-band recording of signals with frequencies ranging up to 125 kHz. The typical sensitivity of the sensors is around -145 dB re 1V/muPa (including preamplifier). Completed in May 2008, AMADEUS consists of six "acoustic clusters", each comprising six acoustic sensors that are arranged at distances of roughly 1 m from each other. Two vertical mechanical structures (so-called lines) of the ANTARES detector host three acoustic clusters each. Spacings between the clusters range from 14.5 to 340 m. Each cluster contains custom-designed electronics boards to amplify and digitise the acoustic signals from the sensors. An on-shore computer cluster is used to process and filter the data stream and store the selected events. The daily volume of recorded data is about 10 GB. The system is operating continuously and automatically, requiring only little human intervention. AMADEUS allows for extensive studies of both transient signals and ambient noise in the deep sea, as well as signal correlations on several length scales and localisation of acoustic point sources. Thus the system is excellently suited to assess the background conditions for the measurement of the bipolar pulses expected to originate from neutrino interactions.
△ Less
Submitted 9 October, 2012; v1 submitted 21 September, 2010;
originally announced September 2010.
-
Study of the interactions of pions in the CALICE silicon-tungsten calorimeter prototype
Authors:
C. Adloff,
Y. Karyotakis,
J. Repond,
J. Yu,
G. Eigen,
Y. Mikami,
N. K. Watson,
J. A. Wilson,
T. Goto,
G. Mavromanolakis,
M. A. Thomson,
D. R. Ward,
W. Yan,
D. Benchekroun,
A. Hoummada,
Y. Khoulaki,
J. Apostolakis,
A. Ribon,
V. Uzhinskiy,
M. Benyamna,
C. Cârloganu,
F. Fehr,
P. Gay,
G. C. Blazey,
D. Chakraborty
, et al. (133 additional authors not shown)
Abstract:
A prototype silicon-tungsten electromagnetic calorimeter for an ILC detector was tested in 2007 at the CERN SPS test beam. Data were collected with electron and hadron beams in the energy range 8 to 80 GeV. The analysis described here focuses on the interactions of pions in the calorimeter. One of the main objectives of the CALICE program is to validate the Monte Carlo tools available for the…
▽ More
A prototype silicon-tungsten electromagnetic calorimeter for an ILC detector was tested in 2007 at the CERN SPS test beam. Data were collected with electron and hadron beams in the energy range 8 to 80 GeV. The analysis described here focuses on the interactions of pions in the calorimeter. One of the main objectives of the CALICE program is to validate the Monte Carlo tools available for the design of a full-sized detector. The interactions of pions in the Si-W calorimeter are therefore confronted with the predictions of various physical models implemented in the GEANT4 simulation framework.
△ Less
Submitted 28 April, 2010;
originally announced April 2010.
-
Calibration systems and methods for the ANTARES neutrino telescope
Authors:
Felix Fehr
Abstract:
The ANTARES neutrino telescope is currently being constructed in the Mediterranean Sea. The complete detector will consist of 12 strings, supplemented by an additional instrumentation line. Nine strings are at present deployed of which five are already connected to the shore and operating. Each string is equipped with 75 Optical Modules (OMs) housing the photomultipliers to detect the Cherenkov…
▽ More
The ANTARES neutrino telescope is currently being constructed in the Mediterranean Sea. The complete detector will consist of 12 strings, supplemented by an additional instrumentation line. Nine strings are at present deployed of which five are already connected to the shore and operating. Each string is equipped with 75 Optical Modules (OMs) housing the photomultipliers to detect the Cherenkov light induced by the charged particles produced in neutrino reactions. An accurate measurement of the Cherenkov photon arrival times as well as the positions and orientations of the OMs is required for a precise reconstruction of the direction of the detected neutrinos. For this purpose the ANTARES detector is provided with several system s to facilitate the calibration of the detector. The time calibration is performed using light pulses emitted from LED and laser devices. The positioning is done via acoustic triangulation using hydrophones. Additionally, local tilt angles and the orientations of the modules are measured with a set of tiltmeters and compasses. In this paper, it is demonstrated that the ANTARES time and alignment calibration systems operate successfully in situ. In particular, it is shown that the ANTARES read-out electronics is capable of reaching a sub-nanosecond time resolution.
△ Less
Submitted 14 November, 2007;
originally announced November 2007.
