The adjoint sensitivity method scalably computes gradients of solutions toordinary differential equations. We generalize this method to stochasticdifferential equations, allowing time-efficient... | Continue reading
Recent huge technological development of Unmanned Aerial Vehicles (UAVs) canprovide breakthrough means of fighting wildland fires. We propose an innovativeforest firefighting system based on the... | Continue reading
Traditional compilers operate on a single generic intermediate representation(IR). These IRs are usually low-level and close to machine instructions. As aresult, optimizations relying on... | Continue reading
This paper seeks to establish a framework for directing a society of simple,specialized, self-interested agents to solve what traditionally are posed asmonolithic single-agent sequential... | Continue reading
The Lottery Ticket Hypothesis is a conjecture that every large neural networkcontains a subnetwork that, when trained in isolation, achieves comparableperformance to the large network. An even... | Continue reading
Quantum mechanics in the Wigner-von Neumann interpretation is presented. Thisis characterized by 1) a quantum dualism between matter and consciousnessunified within an informational neutral... | Continue reading
We study the problem of learning efficient algorithms that stronglygeneralize in the framework of neural program induction. By carefully designingthe input / output interfaces of the neural... | Continue reading
This paper introduces a new task of politeness transfer which involvesconverting non-polite sentences to polite sentences while preserving themeaning. We also provide a dataset of more than 1.39... | Continue reading
We present YOLO, a new approach to object detection. Prior work on objectdetection repurposes classifiers to perform detection. Instead, we frame objectdetection as a regression problem to... | Continue reading
We introduce YOLO9000, a state-of-the-art, real-time object detection systemthat can detect over 9000 object categories. First we propose variousimprovements to the YOLO detection method, both... | Continue reading
OT (Operational Transformation) was invented for supporting real-timeco-editors in the late 1980s and has evolved to become a core technique used intoday's working co-editors and adopted in... | Continue reading
The goal of domain generalization algorithms is to predict well ondistributions different from those seen during training. While a myriad ofdomain generalization algorithms exist,... | Continue reading
We present Breakthrough Listen's "Exotica" Catalog as the centerpiece of ourefforts to expand the diversity of targets surveyed in the Search forExtraterrestrial Intelligence (SETI). As... | Continue reading
We discuss a trajectory of penguins' faeces after the powerful shooting dueto their strong rectal pressure. Practically, it is important to see how farfaeceses reach when penguins expel them... | Continue reading
Haskell, as implemented in the Glasgow Haskell Compiler (GHC), has beenadding new type-level programming features for some time. Many of thesefeatures---chiefly: generalized algebraic datatypes... | Continue reading
We study training of Convolutional Neural Networks (CNNs) with ReLUactivations and introduce exact convex optimization formulations with apolynomial complexity with respect to the number of data... | Continue reading
Non-pharmaceutical interventions (NPIs) have been crucial in curbing COVID-19in the United States (US). Consequently, relaxing NPIs through a phasedre-opening of the US amid still-high levels of... | Continue reading
Duplicate detection is the problem of identifying whether a given item haspreviously appeared in a (possibly infinite) stream of data, when only alimited amount of memory is available. ... | Continue reading
If a semantically open language has no constraints on self-reference, one canshow there is a predicate which is both satisfied and unsatisfied by aself-referring term. The argument requires... | Continue reading
The recent works such as InvisiSpec, SafeSpec, and Cleanup-Spec, amongothers, provided promising solutions to defend speculation induced (transient)attacks. However, they intro-duce delay either... | Continue reading
Message-oriented-middleware (MOM) has become an small industry. MOM offersqueued transaction processing as an advance over pure client-server transactionprocessing. This note makes four points:... | Continue reading
Neural network pruning---the task of reducing the size of a network byremoving parameters---has been the subject of a great deal of work in recentyears. We provide a meta-analysis of the... | Continue reading
Traditional machine instruction level reverse mode automatic differentiation(AD) faces the problem of having a space overhead that linear to time in orderto trace back the computational state,... | Continue reading
Synthetic data is an increasingly popular tool for training deep learningmodels, especially in computer vision but also in other areas. In this work, weattempt to provide a comprehensive survey... | Continue reading
Neural network scaling has been critical for improving the model quality inmany real-world machine learning applications with vast amounts of trainingdata and compute. Although this trend of... | Continue reading
There are countless reasons cited in scientific studies to explain thedifficulties in programming learning. The reasons range from the subject'scomplexity, the ineffective teaching and study... | Continue reading
The conceptual design and flight controller of a novel kind of quadcopter arepresented. This design is capable of morphing the shape of the UAV duringflight to achieve position and attitude... | Continue reading
It is often claimed that the security of quantum key distribution (QKD) isguaranteed by the laws of physics. However, this claim is content-free if theunderlying theoretical definition of QKD is... | Continue reading
Citations are often used as a metric of the impact of scientificpublications. Here, we examine how the number of downloads from Sci-hub as wellas various characteristics of publications and... | Continue reading
We develop a general approach to distill symbolic representations of alearned deep model by introducing strong inductive biases. We focus on GraphNeural Networks (GNNs). The technique works as... | Continue reading
Advances in computing power, natural language processing, and digitization oftext now make it possible to study a culture's evolution through its textsusing a "big data" lens. Our ability to... | Continue reading
Whether enjoying the lucid prose of a favorite author or slogging throughsome other writer's cumbersome, heavy-set prattle (full of parentheses, emdashes, compound adjectives, and Oxford... | Continue reading
A planetary atmosphere is the outer gas layer of a planet. Besides itsscientific significance among the first and most accessible planetary layersobserved from space, it is closely connected... | Continue reading
Black-box machine translation systems have proven incredibly useful for avariety of applications yet by design are hard to adapt, tune to a specificdomain, or build on top of. In this work, we... | Continue reading
Progress in Reinforcement Learning (RL) algorithms goes hand-in-hand with thedevelopment of challenging environments that test the limits of currentmethods. While existing RL environments are... | Continue reading
LiteX is a GitHub-hosted SoC builder / IP library and utilities that can beused to create SoCs and full FPGA designs. Besides being open-source and BSDlicensed, its originality lies in the fact... | Continue reading
Combining the predictions of collections of neural networks often outperformsthe best single network. Such ensembles are typically trained independently,and their superior `wisdom of the crowd'... | Continue reading
Is it possible to automatically classify images without the use ofground-truth annotations? Or when even the classes themselves, are not a prioriknown? These remain important, and open questions... | Continue reading
Deep learning based models have surpassed classical machine learning basedapproaches in various text classification tasks, including sentiment analysis,news categorization, question answering,... | Continue reading
Expert problem-solving is driven by powerful languages for thinking aboutproblems and their solutions. Acquiring expertise means learning theselanguages -- systems of concepts, alongside the... | Continue reading
The vast majority of compact binary mergers in the Universe producegravitational waves that are too weak to yield unambiguous detections; they areunresolved. We present a method to infer the... | Continue reading
Return Oriented Programming (ROP) is a technique by which an attacker caninduce arbitrary behavior inside a vulnerable program without injecting amalicious code. The continues failure of the... | Continue reading
The Variational Auto-Encoder (VAE) is a simple, efficient, and popular deepmaximum likelihood model. Though usage of VAEs is widespread, the derivation ofthe VAE is not as widely understood. In... | Continue reading
Machine learning (ML) models, e.g., deep neural networks (DNNs), arevulnerable to adversarial examples: malicious inputs modified to yielderroneous model outputs, while appearing unmodified to... | Continue reading
One paradigm for learning from few labeled examples while making best use ofa large amount of unlabeled data is unsupervised pretraining followed bysupervised fine-tuning. Although this paradigm... | Continue reading
India accounts for 11% of maternal deaths globally where a woman dies inchildbirth every fifteen minutes. Lack of access to preventive care informationis a significant problem contributing to... | Continue reading
We present a systematic study of biases in natural language generation (NLG)by analyzing text generated from prompts that contain mentions of differentdemographic groups. In this work, we... | Continue reading