Search

Recently, deep reinforcement learning (DRL) methods have achieved impressive performance on tasks in a variety of domains. However, neural network policies produced with DRL methods are not human-interpretable and often have difficulty generalizing to novel scenarios. To address these issues, prior works explore learning programmatic policies that are more interpretable and structured for generalization. Yet, these works either employ limited policy representations (e.g. decision trees, state machines, or predefined program templates) or require stronger supervision (e.g. input/output state pairs or expert demonstrations). We present a framework that instead learns to synthesize a program, which details the procedure to solve a task in a flexible and expressive manner, solely from reward signals. To alleviate the difficulty of learning to compose programs to induce the desired agent behavior from scratch, we propose to first learn a program embedding space that continuously parameterizes diverse behaviors in an unsupervised manner and then search over the learned program embedding space to yield a program that maximizes the return for a given task. Experimental results demonstrate that the proposed framework not only learns to reliably synthesize task-solving programs but also outperforms DRL and program synthesis baselines while producing interpretable and more generalizable policies. We also justify the necessity of the proposed two-stage learning scheme as well as analyze various methods for learning the program embedding., Comment: NeurIPS 2021. 53 pages, 16 figures, 12 tables. Website at https://clvrai.github.io/leaps/

The purpose of this study is to differentiate malignant and benign mediastinal lesions by using the three-dimensional convolutional neural network through the endobronchial ultrasound (EBUS) image. Compared with previous study, our proposed model is robust to noise and able to fuse various imaging features and spatiotemporal features of EBUS videos. Endobronchial ultrasound-guided transbronchial needle aspiration (EBUS-TBNA) is a diagnostic tool for intrathoracic lymph nodes. Physician can observe the characteristics of the lesion using grayscale mode, doppler mode, and elastography during the procedure. To process the EBUS data in the form of a video and appropriately integrate the features of multiple imaging modes, we used a time-series three-dimensional convolutional neural network (3D CNN) to learn the spatiotemporal features and design a variety of architectures to fuse each imaging mode. Our model (Res3D_UDE) took grayscale mode, Doppler mode, and elastography as training data and achieved an accuracy of 82.00% and area under the curve (AUC) of 0.83 on the validation set. Compared with previous study, we directly used videos recorded during procedure as training and validation data, without additional manual selection, which might be easier for clinical application. In addition, model designed with 3D CNN can also effectively learn spatiotemporal features and improve accuracy. In the future, our model may be used to guide physicians to quickly and correctly find the target lesions for slice sampling during the inspection process, reduce the number of slices of benign lesions, and shorten the inspection time.

We study inclusive $J/\psi$ photoproduction at NLO at large $P_T$ at HERA and the EIC. Our computation includes NLO QCD leading-$P_T$ corrections, QED contributions via an off-shell photon as well as those from $J/\psi$+charm channels. For the latter, we employ the variable-flavour-number scheme. Our results are found to agree with the latest HERA data by H1 and provide, for the first time, a reliable estimate of the EIC reach for such a measurement. Finally, we demonstrate the observability of $J/\psi$+charm production and the sensitivy to probe the non-perturbative charm content of the proton at high $x$, also known as intrinsic charm, at the EIC., Comment: 7 pages, 3 figures. Contribution to the XXVIII International Workshop on Deep-Inelastic Scattering and Related Subjects (DIS2021). Submission to SciPost Proceedings. Typo corrected

In this work we compute for the first time the so-called Complete-NLO predictions for top-quark pair hadroproduction in association with at least one isolated photon ($ t \bar t \gamma$). We also compute NLO QCD+EW predictions for the similar case with at least two isolated photons ($ t \bar t \gamma\gamma$) and for single-top hadroproduction in association with at least one isolated photon. In addition, we complement our results with NLO QCD+EW predictions of the hadronic and leptonic decays of top-quark including an isolated photon. All these results have been obtained in a completely automated approach, by extending the capabilities of the MadGraph5_aMC@NLO framework and enabling the Complete-NLO predictions for processes with isolated photons in the final state. We discuss the technical details of the implementation, which involves a mixed EW renormalisation scheme for such processes., Comment: 32 pages, 4 figures, 5 tables. This version matches the one published

We provide a comprehensive discussion, together with a complete setup for simulations, relevant for the production of a single vector-like quark at hadron colliders. Our predictions include finite width effects, signal-background interference effects and next-to-leading order QCD corrections. We explicitly apply the framework to study the single production of a vector-like quark $T$ with charge 2/3, but the same procedure can be used to analyse the single production of vector-like quarks with charge $-4/3$, $-1/3$, $2/3$ and $5/3$, when the vector-like quark interacts with the Standard Model quarks and electroweak bosons. Moreover, this procedure can be straightforwardly extended to include additional interactions with exotic particles. We provide quantitative results for representative benchmark scenarios characterised by the $T$ mass and width, and we determine the role of the interference terms for a range of masses and widths of phenomenological significance. We additionally describe in detail, both analytically and numerically, a striking feature in the invariant mass distribution appearing only in the $T \to th$ channel., Comment: 55 pages, 27 figures, v3:typo in normalization factors Eq.(3.13) fixed, Fig. 4 replaced