168 research outputs found
Event Stitching in the LHCb Software Framework
No full textThis report documents the progress I made on my Summer Student project at CERN during the summer of 2014. A software package for combining particle information from several detector events for analysis purposes is presented. The implementation of the package is discussed and the package is applied to demonstrate a data-driven measurement of the decay time resolution in the decay channel B0s → D−s D+s
root_numpy: 4.3.0
No full text<p>Added support for strings and fixed-size subarrays of basic types in array2tree and array2root.</p>
rootpy/root_numpy: 4.7.1
No full text<p>What's new in this release:</p>
<ul>
<li>Switched from MIT to new BSD license</li>
<li>The branches argument for tree2array and root2array can contain tuples of the form <code>(branch_or_expression, fill_value, length)</code> that allow you to truncate variable-length branches or expressions at a fixed length or single value (if <code>length==1</code> or omitted) while imputing missing values where the original subarray was shorter than <code>length</code> with <code>fill_value</code>.</li>
</ul>
scikit-hep/root_numpy: 4.7.2
No full textWhat's new in this release:
<ul>
<li>The tests were missing from 4.7.1 and have been re-included</li>
</ul>
Data for "Inverse Scaling: When Bigger Isn't Better"
No full textWork on scaling laws has found that large language models (LMs) show predictable improvements to overall loss with increased scale (model size, training data, and compute). Here, we present evidence for the claim that LMs may show inverse scaling, or worse task performance with increased scale, e.g., due to flaws in the training objective and data. We present empirical evidence of inverse scaling on 11 datasets collected by running a public contest, the Inverse Scaling Prize, with a substantial prize pool. Through analysis of the datasets, along with other examples found in the literature, we identify four potential causes of inverse scaling: (i) preference to repeat memorized sequences over following in-context instructions, (ii) imitation of undesirable patterns in the training data, (iii) tasks containing an easy distractor task which LMs could focus on, rather than the harder real task, and (iv) correct but misleading few-shot demonstrations of the task. We release the winning datasets at https://github.com/inverse-scaling/prize/tree/main/data-release to allow for further investigation of inverse scaling. Our tasks have helped drive the discovery of U-shaped and inverted-U scaling trends, where an initial trend reverses, suggesting that scaling trends are less reliable at predicting the behavior of larger-scale models than previously understood. Overall, our results suggest that there are tasks for which increased model scale alone may not lead to progress, and that more careful thought needs to go into the data and objectives for training language models.We thank everyone who submitted tasks to the Inverse Scaling Prize. Thank you to all the volunteers who
contributed to reviewing submissions: Ananya Harsh Jha, Beth Barnes, Jonas Pfeiffer, Joshua Landau,
Kamile Lukosiute, Naomi Saphra, Nicholas Kees Dupuis, Nicholas Lourie, Peter Barnett, Quintin Pope,
Rasika Bhalerao, Richard Pang, Rune Kvist, Sam Ringer, Tamera Lanham, Thomas Larsen, and William
Merrill.
We are grateful to Open Philanthropy for providing funding for the prize. Thanks to Hannah Betts, Karl
Berzins, Josh Jacobson, and Adam Gleave from FAR AI for logistical support in all aspects of handling prize
money, including funding applications and distributing prizes. Thanks to Mary Dowling and Julie Nguyen
from Tovella Dowling. Thanks also to Jenna Webster, Andrew Morton, and Brandon Warehime from Players
Philanthropy Fund.
This project has benefited from financial support to SB by Eric and Wendy Schmidt (made by recommendation of the Schmidt Futures program) and Open Philanthropy, and from in-kind support by the NYU
High-Performance Computing Center and Stability AI. This material is based upon work supported by the
National Science Foundation under Grant Nos. 1922658 and 2046556. Any opinions, findings, and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect
the views of the National Science Foundation.
We would like to thank Anthropic for the use of their LMs, OpenAI for API help and credits for participants,
including Cameron McKinnon for help evaluating on Anthropic models. We would also like to thank Scott
Heiner, Edwin Chen, and others from Surge AI for organizing human validation and offering support to
participants, and Jason Phang, Stella Biderman, and HuggingFace for their help running evaluations on
large public models.
Thanks to Lama Ahmad and others from OpenAI for assistance to participants in running evaluations on
the OpenAI API, and for providing API credits. We also thank Ilya Sutskever and others at OpenAI for
sharing results on GPT-4 models.
We thank DeepMind for running evaluations, in particular Matthew Rahtz for his work running evaluations
on Gopher and Chinchilla in both rounds and for his quick turnaround and patience in re-running after data
issues.
From DeepMind, we also thank Nick Fernando, Sanah Choudhry, and Koray Kavukcuoglu, and the teams
behind Gopher (Jack W. Rae, Sebastian Borgeaud, Trevor Cai, Katie Millican, Jordan Hoffmann, Francis
Song, John Aslanides, Sarah Henderson, Roman Ring, Susannah Young, Eliza Rutherford, Tom Hennigan,
Jacob Menick, Albin Cassirer, Richard Powell, George van den Driessche, Lisa Anne Hendricks, Maribeth
Rauh, Po-Sen Huang, Amelia Glaese, Johannes Welbl, Sumanth Dathathri, Saffron Huang, Jonathan Uesato,
John Mellor, Irina Higgins, Antonia Creswell, Nat McAleese, Amy Wu, Erich Elsen, Siddhant Jayakumar,
Elena Buchatskaya, David Budden, Esme Sutherland, Karen Simonyan, Michela Paganini, Laurent Sifre,
Lena Martens, Xiang Lorraine Li, Adhiguna Kuncoro, Aida Nematzadeh, Elena Gribovskaya, Domenic Donato, Angeliki Lazaridou, Arthur Mensch, Jean-Baptiste Lespiau, Maria Tsimpoukelli, Nikolai Grigorev,
Doug Fritz, Thibault Sottiaux, Mantas Pajarskas, Toby Pohlen, Zhitao Gong, Daniel Toyama, Cyprien de
Masson d’Autume, Yujia Li, Tayfun Terzi, Vladimir Mikulik, Igor Babuschkin, Aidan Clark, Diego de Las
Casas, Aurelia Guy, Chris Jones, James Bradbury, Matthew Johnson, Blake Hechtman, Laura Weidinger,
Iason Gabriel, William Isaac, Ed Lockhart, Simon Osindero, Laura Rimell, Chris Dyer, Oriol Vinyals, Kareem Ayoub, Jeff Stanway, Lorrayne Bennett, Demis Hassabis, Koray Kavukcuoglu, Geoffrey Irving) and
Chinchilla (Jordan Hoffmann, Sebastian Borgeaud, Arthur Mensch, Elena Buchatskaya, Trevor Cai, Eliza
Rutherford, Diego de Las Casas, Lisa Anne Hendricks, Johannes Welbl, Aidan Clark, Tom Hennigan, Eric
Noland, Katie Millican, George van den Driessche, Bogdan Damoc, Aurelia Guy, Simon Osindero, Karen
Simonyan, Erich Elsen, Jack W. Rae, Oriol Vinyals, Laurent Sifre
scikit-hep/root_numpy: 4.8.0
No full textWhat's new in this release:
<ul>
<li>coverage fixes and new build matrix for ROOT for CI</li>
<li>support python 3.7 (with new cython)</li>
<li>fix typos in docs</li>
</ul>
Studies of the resonance structure in D0→K∓π±π±π∓D0→K∓π±π±π∓ decays
No full textAmplitude models are constructed to describe the resonance structure of D0→K−π+π+π−D0→K−π+π+π− and D0→K+π−π−π+D0→K+π−π−π+ decays using pp collision data collected at centre-of-mass energies of 7 and 8 TeV with the LHCb experiment, corresponding to an integrated luminosity of 3.0 fb−1fb−1 . The largest contributions to both decay amplitudes are found to come from axial resonances, with decay modes D0→a1(1260)+K−D0→a1(1260)+K− and D0→K1(1270/1400)+π−D0→K1(1270/1400)+π− being prominent in D0→K−π+π+π−D0→K−π+π+π− and D0→K+π−π−π+D0→K+π−π−π+ , respectively. Precise measurements of the lineshape parameters and couplings of the a1(1260)+a1(1260)+ , K1(1270)−K1(1270)− and K(1460)−K(1460)− resonances are made, and a quasi model-independent study of the K(1460)−K(1460)− resonance is performed. The coherence factor of the decays is calculated from the amplitude models to be RK3π=0.459±0.010(stat)±0.012(syst)±0.020(model)RK3π=0.459±0.010(stat)±0.012(syst)±0.020(model) , which is consistent with direct measurements. These models will be useful in future measurements of the unitary-triangle angle γγ and studies of charm mixing and CPCP violation
A measurement of the CP asymmetry difference between Λ c + → pK−K+ and pπ−π+ decays
No full textThe difference between the CP asymmetries in the decays Λ c + → pK −K + and Λ c + → pπ −π + is presented. Proton-proton collision data taken at centre-of-mass energies of 7 and 8 TeV collected by the LHCb detector in 2011 and 2012 are used, corresponding to an integrated luminosity of 3 fb −1. The Λ c + candidates are reconstructed as part of the Λ b 0 → Λ c +μ −X decay chain. In order to maximize the cancellation of production and detection asymmetries in the difference, the final-state kinematic distributions of the two samples are aligned by applying phase-space-dependent weights to the Λ c + → pπ −π + sample. This alters the definition of the integrated CP asymmetry to A CP wgt(pπ −π +). Both samples are corrected for reconstruction and selection efficiencies across the five-dimensional Λ c + decay phase space. The difference in CP asymmetries is found to be ΔACPwgt=ACP(pK−K+)−ACPwgt(pπ−π+)=(0.30±0.91±0.61)%, where the first uncertainty is statistical and the second is systematic.[Figure not available: see fulltext.].</p
Search for the rare decay Λ+ c → pμ + μ −
No full textA search for the flavor-changing neutral-current decay Λ + c → p μ + μ − is reported using a data set corresponding to an integrated luminosity of 3.0 fb − 1 collected by the LHCb Collaboration. No significant signal is observed outside of the dimuon mass regions around the ϕ and ω resonances, and an upper limit is placed on the branching fraction of B ( Λ + c → p μ + μ − ) < 7.7 ( 9.6 ) × 10 − 8 at 90%(95%) confidence level. A significant signal is observed in the ω dimuon mass region for the first time
Search for B-c(+) decays to two charm mesons
No full textA search for decays of B-c(+) mesons to two charm mesons is performed for the first time using data corresponding to an integrated luminosity of 3.0 fb(-1), collected by the LHCb experiment in pp collisions at centre-of-mass energies of 7 and 8 TeV. The decays considered are B-c(+)-> D-(s)(()*())(+) (D) over bar (()*()0) and Bc(+)-> D-(s)(()*D-)+(()*())(0), which are normalised to high-yield B+-> D-(s)(+)(D) over bar (0)decays. No evidence for a signal is found and limits are set on twelve B-c(+) decay modes. (C) 2018 The Author(s). Published by Elsevier B.V
- …
