Search for: All records

The inductive biases of trained neural networks are difficult to understand and, consequently, to adapt to new settings. We study the inductive biases of linearizations of neural networks, which we show to be surprisingly good summaries of the full network functions. Inspired by this finding, we propose a technique for embedding these inductive biases into Gaussian processes through a kernel designed from the Jacobian of the network. In this setting, domain adaptation takes the form of interpretable posterior inference, with accompanying uncertainty estimation. This inference is analytic and free of local optima issues found in standard techniques such as fine-tuning neural network weights to a new task. We develop significant computational speed-ups based on matrix multiplies, including a novel implementation for scalable Fisher vector products. Our experiments on both image classification and regression demonstrate the promise and convenience of this framework for transfer learning, compared to neural network fine-tuning. 

A bstract We report on a measurement of the $$ {\Lambda}_c^{+} $$ Λ c + to D 0 production ratio in peripheral PbPb collisions at $$ \sqrt{s_{\textrm{NN}}} $$ s NN = 5 . 02 TeV with the LHCb detector in the forward rapidity region 2 < y < 4 . 5. The $$ {\Lambda}_c^{+} $$ Λ c + ( D 0 ) hadrons are reconstructed via the decay channel $$ {\Lambda}_c^{+} $$ Λ c + → pK − π + ( D 0 → K − π + ) for 2 < p T < 8 GeV/ c and in the centrality range of about 65–90%. The results show no significant dependence on p T , y or the mean number of participating nucleons. They are also consistent with similar measurements obtained by the LHCb collaboration in pPb and Pbp collisions at $$ \sqrt{s_{\textrm{NN}}} $$ s NN = 5 . 02 TeV. The data agree well with predictions from PYTHIA in pp collisions at $$ \sqrt{s} $$ s = 5 TeV but are in tension with predictions of the Statistical Hadronization model. 

A bstract A search for the lepton-flavour violating decays B 0 → K *0 μ ± e ∓ and $$ {B}_s^0 $$ B s 0 → ϕμ ± e ∓ is presented, using proton-proton collision data collected by the LHCb detector at the LHC, corresponding to an integrated luminosity of 9 fb − 1 . No significant signals are observed and upper limits of $$ {\displaystyle \begin{array}{c}\mathcal{B}\left({B}^0\to {K}^{\ast 0}{\mu}^{+}{e}^{-}\right)<5.7\times {10}^{-9}\left(6.9\times {10}^{-9}\right),\\ {}\mathcal{B}\left({B}^0\to {K}^{\ast 0}{\mu}^{-}{e}^{+}\right)<6.8\times {10}^{-9}\left(7.9\times {10}^{-9}\right),\\ {}\mathcal{B}\left({B}^0\to {K}^{\ast 0}{\mu}^{\pm }{e}^{\mp}\right)<10.1\times {10}^{-9}\left(11.7\times {10}^{-9}\right),\\ {}\mathcal{B}\left({B}_s^0\to \phi {\mu}^{\pm }{e}^{\mp}\right)<16.0\times {10}^{-9}\left(19.8\times {10}^{-9}\right)\end{array}} $$ B B 0 → K ∗ 0 μ + e − < 5.7 × 10 − 9 6.9 × 10 − 9 , B B 0 → K ∗ 0 μ − e + < 6.8 × 10 − 9 7.9 × 10 − 9 , B B 0 → K ∗ 0 μ ± e ∓ < 10.1 × 10 − 9 11.7 × 10 − 9 , B B s 0 → ϕ μ ± e ∓ < 16.0 × 10 − 9 19.8 × 10 − 9 are set at 90% (95%) confidence level. These results constitute the world’s most stringent limits to date, with the limit on the decay $$ {B}_s^0 $$ B s 0 → ϕμ ± e ∓ the first being set. In addition, limits are reported for scalar and left-handed lepton-flavour violating New Physics scenarios.