arXiv subjects

David Weinberger

Publications and source records attributed to David Weinberger.

4 recordsLinked to original sources

arXiv2024

Gemma 2: Improving Open Language Models at a Practical Size

In this work, we introduce Gemma 2, a new addition to the Gemma family of lightweight, state-of-the-art open models, ranging in scale from 2 billion to 27 billion parameters. In this new version, we apply several known technical modifications to the Transformer architecture, such as interleaving local-global attentions (Beltagy et al., 2020a) and group-query attention (Ainslie et al., 2023). We also train the 2B and 9B models with knowledge distillation (Hinton et al., 2015) instead of next token prediction. The resulting models deliver the best performance for their size, and even offer competitive alternatives to models that are 2-3 times bigger. We release all our models to the community.

cs.CL↗

arXiv2022

Study of a possible silicon photomultiplier based readout of the large plastic scintillator neutron detector NeuLAND

The NeuLAND (New Large-Area Neutron Detector) plastic-scintillator-based time-of-flight detector for 0.1-1.6 GeV neutrons is currently under construction at the Facility for Antiproton and Ion Research (FAIR), Darmstadt, Germany. In its final configuration, NeuLAND will consist of 3000 2.7 m $\times$ 5 cm $\times$ 5 cm big plastic scintillator bars that are read out on each end by fast timing photomultipliers. Here, data from a comprehensive study of an alternative light readout scheme using silicon photomultipliers (SiPM) are reported. For this purpose, a NeuLAND bar was instrumented on each end with a SiPM-based prototype of the same geometry as a 1'' photomultiplier tube, including four 6 $\times$ 6 mm$^2$ SiPMs, amplifiers, high voltage supply, and microcontroller. Tests were carried out using the 35 MeV electron beam from the superconducting Electron Linac for beams with high Brilliance and low Emittance (ELBE) with its picosecond-level time jitter in two different modes of operation, namely parasitic mode with one electron per bunch and single-user mode with 1-60 electrons per bunch. Acqiris fast digitisers were used for data acquisition. In addition, off-beam tests using cosmic rays and the NeuLAND data acquisition scheme have been carried out. Typical time resolutions of $σ_t\leq$ 120 ps were found for $\geq$95% efficiency for minimum ionising particles, improving on previous work at ELBE and exceeding the NeuLAND timing goal of $σ_t$ < 150 ps. Over a range of 10-300 MeV deposited energy in the NeuLAND bar, the gain was found to deviate by $\leq$10% ($\leq$20%) from linearity for 35 mm (75 mm) SiPM pitch, respectively, satisfactory for calorimetric use of the full NeuLAND detector. The dark rate of the prototype studied was found to be lower than the expected cosmic-ray induced background in NeuLAND.

physics.ins-det↗

arXiv2019

Accountability of AI Under the Law: The Role of Explanation

The ubiquity of systems using artificial intelligence or "AI" has brought increasing attention to how those systems should be regulated. The choice of how to regulate AI systems will require care. AI systems have the potential to synthesize large amounts of data, allowing for greater levels of personalization and precision than ever before---applications range from clinical decision support to autonomous driving and predictive policing. That said, there exist legitimate concerns about the intentional and unintentional negative consequences of AI systems. There are many ways to hold AI systems accountable. In this work, we focus on one: explanation. Questions about a legal right to explanation from AI systems was recently debated in the EU General Data Protection Regulation, and thus thinking carefully about when and how explanation from AI systems might improve accountability is timely. In this work, we review contexts in which explanation is currently required under the law, and then list the technical considerations that must be considered if we desired AI systems that could provide kinds of explanations that are currently required of humans.

cs.AI↗

arXiv2016

Silicon photomultiplier readout of a monolithic 270$\times$5$\times$5 cm$^3$ plastic scintillator bar for time of flight applications

The detection of 200-1000 MeV neutrons requires large amounts, $\sim$100 cm, of detector material because of the long nuclear interaction length of these particles. In the example of the NeuLAND neutron time-of-flight detector at FAIR, this is accomplished by using 3000 monolithic scintillator bars of 270$\times$5$\times$5 cm$^3$ size made of a fast plastic. Each bar is read out on the two long ends, and the needed time resolution of $σ_t$ $<$ 150 ps is reached with fast timing photomultipliers. In the present work, it is investigated whether silicon photomultiplier (SiPM) photosensors can be used instead. Experiments with a picosecond laser system were conducted to determine the timing response of the assembly made up of SiPM and preamplifier. The response of the full system including also the scintillator was studied using 30 MeV single electrons provided by the ELBE superconducting electron linac. The ELBE data were matched by a simple Monte Carlo simulation, and they were found to obey an inverse-square-root scaling law. In the electron beam tests, a time resolution of $σ_t$ = 136 ps was reached with a pure SiPM readout, well within the design parameters for NeuLAND.

physics.ins-det↗