-
OpenAssistant Conversations -- Democratizing Large Language Model Alignment
Authors:
Andreas Köpf,
Yannic Kilcher,
Dimitri von Rütte,
Sotiris Anagnostidis,
Zhi-Rui Tam,
Keith Stevens,
Abdullah Barhoum,
Nguyen Minh Duc,
Oliver Stanley,
Richárd Nagyfi,
Shahul ES,
Sameer Suri,
David Glushkov,
Arnav Dantuluri,
Andrew Maguire,
Christoph Schuhmann,
Huu Nguyen,
Alexander Mattick
Abstract:
Aligning large language models (LLMs) with human preferences has proven to drastically improve usability and has driven rapid adoption as demonstrated by ChatGPT. Alignment techniques such as supervised fine-tuning (SFT) and reinforcement learning from human feedback (RLHF) greatly reduce the required skill and domain knowledge to effectively harness the capabilities of LLMs, increasing their acce…
▽ More
Aligning large language models (LLMs) with human preferences has proven to drastically improve usability and has driven rapid adoption as demonstrated by ChatGPT. Alignment techniques such as supervised fine-tuning (SFT) and reinforcement learning from human feedback (RLHF) greatly reduce the required skill and domain knowledge to effectively harness the capabilities of LLMs, increasing their accessibility and utility across various domains. However, state-of-the-art alignment techniques like RLHF rely on high-quality human feedback data, which is expensive to create and often remains proprietary. In an effort to democratize research on large-scale alignment, we release OpenAssistant Conversations, a human-generated, human-annotated assistant-style conversation corpus consisting of 161,443 messages in 35 different languages, annotated with 461,292 quality ratings, resulting in over 10,000 complete and fully annotated conversation trees. The corpus is a product of a worldwide crowd-sourcing effort involving over 13,500 volunteers. Models trained on OpenAssistant Conversations show consistent improvements on standard benchmarks over respective base models. We release our code and data under a fully permissive licence.
△ Less
Submitted 31 October, 2023; v1 submitted 14 April, 2023;
originally announced April 2023.
-
Physically Meaningful Uncertainty Quantification in Probabilistic Wind Turbine Power Curve Models as a Damage Sensitive Feature
Authors:
J. H. Mclean,
M. R. Jones,
B. J. O'Connell,
A. E Maguire,
T. J. Rogers
Abstract:
A wind turbines' power curve is easily accessible damage sensitive data, and as such is a key part of structural health monitoring in wind turbines. Power curve models can be constructed in a number of ways, but the authors argue that probabilistic methods carry inherent benefits in this use case, such as uncertainty quantification and allowing uncertainty propagation analysis. Many probabilistic…
▽ More
A wind turbines' power curve is easily accessible damage sensitive data, and as such is a key part of structural health monitoring in wind turbines. Power curve models can be constructed in a number of ways, but the authors argue that probabilistic methods carry inherent benefits in this use case, such as uncertainty quantification and allowing uncertainty propagation analysis. Many probabilistic power curve models have a key limitation in that they are not physically meaningful - they return mean and uncertainty predictions outside of what is physically possible (the maximum and minimum power outputs of the wind turbine). This paper investigates the use of two bounded Gaussian Processes in order to produce physically meaningful probabilistic power curve models. The first model investigated was a warped heteroscedastic Gaussian process, and was found to be ineffective due to specific shortcomings of the Gaussian Process in relation to the war** function. The second model - an approximated Gaussian Process with a Beta likelihood was highly successful and demonstrated that a working bounded probabilistic model results in better predictive uncertainty than a corresponding unbounded one without meaningful loss in predictive accuracy. Such a bounded model thus offers increased accuracy for performance monitoring and increased operator confidence in the model due to guaranteed physical plausibility.
△ Less
Submitted 30 September, 2022;
originally announced September 2022.
-
Universal pulses for homogeneous excitation using single channel coils
Authors:
Ronald Mooiweer,
Ian A. Clark,
Eleanor A. Maguire,
Martina F. Callaghan,
Jospeh V. Hajnal,
Shaihan J. Malik
Abstract:
Purpose: Universal Pulses (UPs) are excitation pulses that reduce the flip angle inhomogeneity in high field MRI systems without subject-specific optimization, originally developed for parallel transmit (PTX) systems at 7T. We investigated the potential benefits of UPs for single channel (SC) transmit systems at 3T, which are widely used for clinical and research imaging, and for which flip angle…
▽ More
Purpose: Universal Pulses (UPs) are excitation pulses that reduce the flip angle inhomogeneity in high field MRI systems without subject-specific optimization, originally developed for parallel transmit (PTX) systems at 7T. We investigated the potential benefits of UPs for single channel (SC) transmit systems at 3T, which are widely used for clinical and research imaging, and for which flip angle inhomogeneity can still be problematic.
Methods: SC-UPs were designed using a spiral nonselective k-space trajectory for brain imaging at 3T using transmit field maps (B1+) and off-resonance maps (B0) acquired on two different scanner types: a 'standard' single channel transmit system and a system with a PTX body coil. The effect of training group size was investigated using data (200 subjects) from the standard system. The PTX system was used to compare SC-UPs to PTX-UPs (15 subjects). In two additional subjects, prospective imaging using SC-UP was studied.
Results: Average flip angle error fell from 9.5+/-0.5% for 'default' excitation to 3.0+/-0.6% using SC-UPs trained over 50 subjects. Performance of the UPs was found to steadily improve as training group size increased, but stabilized after ~15 subjects. On the PTX-enabled system, SC-UPs again outperformed default excitation in simulations (4.8+/-0.6% error versus 10.6+/-0.8% respectively) though greater homogenization could be achieved with PTX-UPs (3.9+/-0.6%) and personalized pulses (SC-PP 3.6+/-1.0%, PTX-PP 2.9+/-0.6%). MP-RAGE imaging using SC-UP resulted in greater separation between grey and white matter signal intensities than default excitation.
Conclusions: SC-UPs can improve excitation homogeneity in standard 3T systems without further calibration and could be used instead of a default excitation pulse for nonselective neuroimaging at 3T.
△ Less
Submitted 6 January, 2022;
originally announced January 2022.
-
Bayesian Modelling of Multivalued Power Curves from an Operational Wind Farm
Authors:
L. A. Bull,
P. A. Gardner,
T. J. Rogers,
N. Dervilis,
E. J. Cross,
E. Papatheou,
A. E. Maguire,
C. Campos,
K. Worden
Abstract:
Power curves capture the relationship between wind speed and output power for a specific wind turbine. Accurate regression models of this function prove useful in monitoring, maintenance, design, and planning. In practice, however, the measurements do not always correspond to the ideal curve: power curtailments will appear as (additional) functional components. Such multivalued relationships canno…
▽ More
Power curves capture the relationship between wind speed and output power for a specific wind turbine. Accurate regression models of this function prove useful in monitoring, maintenance, design, and planning. In practice, however, the measurements do not always correspond to the ideal curve: power curtailments will appear as (additional) functional components. Such multivalued relationships cannot be modelled by conventional regression, and the associated data are usually removed during pre-processing. The current work suggests an alternative method to infer multivalued relationships in curtailed power data. Using a population-based approach, an overlap** mixture of probabilistic regression models is applied to signals recorded from turbines within an operational wind farm. The model is shown to provide an accurate representation of practical power data across the population.
△ Less
Submitted 30 November, 2021;
originally announced November 2021.
-
Interference suppression techniques for OPM-based MEG: Opportunities and challenges
Authors:
Robert A Seymour,
Nicholas Alexander,
Stephanie Mellor,
George C O'Neill,
Tim M Tierney,
Gareth R Barnes,
Eleanor A Maguire
Abstract:
One of the primary technical challenges facing magnetoencephalography (MEG) is that the magnitude of neuromagnetic fields is several orders of magnitude lower than interfering signals. Recently, a new type of sensor has been developed - the optically pumped magnetometer (OPM). These sensors can be placed directly on the scalp and move with the head during participant movement, making them wearable…
▽ More
One of the primary technical challenges facing magnetoencephalography (MEG) is that the magnitude of neuromagnetic fields is several orders of magnitude lower than interfering signals. Recently, a new type of sensor has been developed - the optically pumped magnetometer (OPM). These sensors can be placed directly on the scalp and move with the head during participant movement, making them wearable. This opens up a range of exciting experimental and clinical opportunities for OPM-based MEG experiments, including paediatric studies, and the incorporation of naturalistic movements into neuroimaging paradigms. However, OPMs face some unique challenges in terms of interference suppression, especially in situations involving mobile participants, and when OPMs are integrated with electrical equipment required for naturalistic paradigms, such as motion capture systems. Here we briefly review various hardware solutions for OPM interference suppression. We then outline several signal processing strategies aimed at increasing the signal from neuromagnetic sources. These include regression-based strategies, temporal filtering and spatial filtering approaches. The focus is on the practical application of these signal processing algorithms to OPM data. In a similar vein, we include two worked-through experiments using OPM data collected from a whole-head sensor array. These tutorial-style examples illustrate how the steps for suppressing external interference can be implemented, including the associated data and code so that researchers can try the pipelines for themselves. With the popularity of OPM-based MEG rising, there will be an increasing need to deal with interference suppression. We hope this practical paper provides a resource for OPM-based MEG researchers to build upon.
△ Less
Submitted 29 November, 2021; v1 submitted 6 October, 2021;
originally announced October 2021.
-
First Results from the REAL-time Transient Acquisition backend (REALTA) at the Irish LOFAR station
Authors:
P. C. Murphy,
P. Callanan,
J. McCauley,
D. J. McKenna,
D. Ó Fionnagáin,
C. K. Louis,
M. P. Redman,
L. A. Cañizares,
E. P. Carley,
S. A. Maloney,
B. Coghlan,
M. Daly,
J. Scully,
J. Dooley,
V. Gajjar,
C. Giese,
A. Brennan,
E. F. Keane,
C. A. Maguire,
J. Quinn,
S. Mooney,
A. M. Ryan,
J. Walsh,
C. M. Jackman,
A. Golden
, et al. (5 additional authors not shown)
Abstract:
Modern radio interferometers such as the LOw Frequency ARray (LOFAR) are capable of producing data at hundreds of gigabits to terabits per second. This high data rate makes the analysis of radio data cumbersome and computationally expensive. While high performance computing facilities exist for large national and international facilities, that may not be the case for instruments operated by a sing…
▽ More
Modern radio interferometers such as the LOw Frequency ARray (LOFAR) are capable of producing data at hundreds of gigabits to terabits per second. This high data rate makes the analysis of radio data cumbersome and computationally expensive. While high performance computing facilities exist for large national and international facilities, that may not be the case for instruments operated by a single institution or a small consortium. Data rates for next generation radio telescopes are set to eclipse those currently in operation, hence local processing of data will become all the more important. Here, we introduce the REAL-time Transient Acquisition backend (REALTA), a computing backend at the Irish LOFAR station (I-LOFAR) which facilitates the recording of data in near real-time and post-processing. We also present first searches and scientific results of a number of radio phenomena observed by I-LOFAR and REALTA, including pulsars, fast radio bursts (FRBs), rotating radio transients (RRATs), the search for extraterrestrial intelligence (SETI), Jupiter, and the Sun.
△ Less
Submitted 25 August, 2021;
originally announced August 2021.
-
LOFAR observations of a jet-driven piston shock in the low solar corona
Authors:
Ciara A. Maguire,
Eoin P. Carley,
Pietro Zucca,
Nicole Vilmer,
Peter T. Gallagher
Abstract:
The Sun produces highly dynamic and eruptive events that can drive shocks through the corona. These shocks can accelerate electrons, which result in plasma emission in the form of a type II radio burst. Despite the large number of type II radio bursts observations, the precise origin of coronal shocks is still subject to investigation. Here we present a well observed solar eruptive event that occu…
▽ More
The Sun produces highly dynamic and eruptive events that can drive shocks through the corona. These shocks can accelerate electrons, which result in plasma emission in the form of a type II radio burst. Despite the large number of type II radio bursts observations, the precise origin of coronal shocks is still subject to investigation. Here we present a well observed solar eruptive event that occurred on 16 October 2015, focusing on a jet observed in the extreme ultraviolet (EUV) by the Atmospheric Imaging Assembly (SDO/AIA), a streamer observed in white-light by the Large Angle and Spectrometric Coronagraph (SOHO/LASCO), and a metric type II radio burst observed by the LOw Frequency Array (LOFAR). LOFAR interferometrically imaged the fundamental and harmonic sources of a type II radio burst and revealed that the sources did not appear to be co-spatial, as would be expected from the plasma emission mechanism. We correct for the separation between the fundamental and harmonic using a model which accounts for scattering of radio waves by electron density fluctuations in a turbulent plasma. This allows us to show the type II radio sources were located $\sim$0.5 R$_\odot$ above the jet and propagated at a speed of $\sim$1000 kms$^{-1}$, which was significantly faster than the jet speed of $\sim$200 kms$^{-1}$. This suggests that the type II burst was generated by a piston shock driven by the jet in the low corona.
△ Less
Submitted 14 January, 2021;
originally announced January 2021.
-
Evolution of the Alfvén Mach number associated with a coronal mass ejection shock
Authors:
Ciara A. Maguire,
Eoin P. Carley,
Joseph McCauley,
Peter T. Gallagher
Abstract:
The Sun regularly produces large-scale eruptive events, such as coronal mass ejections (CMEs) that can drive shock waves through the solar corona. Such shocks can result in electron acceleration and subsequent radio emission in the form of a type II radio burst. However, the early-phase evolution of shock properties and its relationship to type II burst evolution is still subject to investigation.…
▽ More
The Sun regularly produces large-scale eruptive events, such as coronal mass ejections (CMEs) that can drive shock waves through the solar corona. Such shocks can result in electron acceleration and subsequent radio emission in the form of a type II radio burst. However, the early-phase evolution of shock properties and its relationship to type II burst evolution is still subject to investigation. Here we study the evolution of a CME-driven shock by comparing three commonly used methods of calculating the Alfvén Mach number ($M_A$), namely: shock geometry, a comparison of CME speed to a model of the coronal Alfvén speed, and the type II band-splitting method. We applied the three methods to the 2017 September 2 event, focusing on the shock wave observed in extreme ultraviolet (EUV) by the Solar Ultraviolet Imager (SUVI) on board GOES-16, in white-light by the Large Angle and Spectrometric Coronagraph (LASCO) on board SOHO, and the type II radio burst observed by the Irish Low Frequency Array (I-LOFAR). We show that the three different methods of estimating shock $M_A$ yield consistent results and provide a means of relating shock property evolution to the type II emission duration. The type II radio emission emerged from near the nose of the CME when $M_A$ was in the range 1.4-2.4 at a heliocentric distance of $\sim$1.6 $R_\odot$. The emission ceased when the CME nose reached $\sim$2.4 $R_\odot$, despite an increasing Alfvén Mach number (up to 4). We suggest the radio emission cessation is due to the lack of quasi-perpendicular geometry at this altitude, which inhibits efficient electron acceleration and subsequent radio emission.
△ Less
Submitted 6 December, 2019; v1 submitted 4 December, 2019;
originally announced December 2019.
-
Simulations of an offshore wind farm using large eddy simulation and a torque-controlled actuator disc model
Authors:
Angus Creech,
Wolf-Gerrit Früh,
A. Eoghan Maguire
Abstract:
We present here a computational fluid dynamics (CFD) simulation of Lillgrund offshore wind farm, which is located in the Oresund Strait between Sweden and Denmark. The simulation combines a dynamic representation of wind turbines embedded within a Large-Eddy Simulation CFD solver, and uses hr-adaptive meshing to increase or decrease mesh resolution where required. This allows the resolution of bot…
▽ More
We present here a computational fluid dynamics (CFD) simulation of Lillgrund offshore wind farm, which is located in the Oresund Strait between Sweden and Denmark. The simulation combines a dynamic representation of wind turbines embedded within a Large-Eddy Simulation CFD solver, and uses hr-adaptive meshing to increase or decrease mesh resolution where required. This allows the resolution of both large scale flow structures around the wind farm, and local flow conditions at individual turbines; consequently, the response of each turbine to local conditions can be modelled, as well as the resulting evolution of the turbine wakes. This paper provides a detailed description of the turbine model which simulates interactions between the wind, turbine rotors, and turbine generators by calculating the forces on the rotor, the body forces on the air, and instantaneous power output. This model was used to investigate a selection of key wind speeds and directions, investigating cases where a row of turbines would be aligned with the wind or at specific angles to the wind. Results shown include presentations of the spin-up of turbines, the observation of eddies moving through the turbine array, meandering turbine wakes, and an extensive wind farm wake several kilometres in length. The key measurement available for cross-validation with operational wind farm data is the power output from the individual turbines, where the effect of unsteady turbine wakes on the performance of downstream turbines was a point of interest. The results from simulations were compared to performance measurements from the real wind farm to provide a firm quantitative validation of this methodology. Having achieved good agreement between the model and actual wind farm measurements, the potential of the methodology to provide a tool for further investigations of engineering and atmospheric science problems is outlined.
△ Less
Submitted 13 December, 2014; v1 submitted 14 October, 2014;
originally announced October 2014.
