Robot localization remains a challenging task in GPS denied environments. State estimation approaches based on local sensors, e.g. cameras or IMUs, are drifting-prone for long-range missions as error accumulates. In this study, we aim to address this
problem by localizing image observations in a 2D multi-modal geospatial map. We introduce the cross-scale dataset and a methodology to produce additional data from cross-modality sources. We propose a framework that learns cross-scale visual representations without supervision. Experiments are conducted on data from two different domains, underwater and aerial. In contrast to existing studies in cross-view image geo-localization, our approach a) performs better on smaller-scale multi-modal maps; b) is more computationally efficient for real-time applications; c) can serve directly in concert with state estimation pipelines.
We study how masking and predicting tokens in an unsupervised fashion can give rise to linguistic structures and downstream performance gains. Recent theories have suggested that pretrained language models acquire useful inductive biases through mask
s that implicitly act as cloze reductions for downstream tasks. While appealing, we show that the success of the random masking strategy used in practice cannot be explained by such cloze-like masks alone. We construct cloze-like masks using task-specific lexicons for three different classification datasets and show that the majority of pretrained performance gains come from generic masks that are not associated with the lexicon. To explain the empirical success of these generic masks, we demonstrate a correspondence between the Masked Language Model (MLM) objective and existing methods for learning statistical dependencies in graphical models. Using this, we derive a method for extracting these learned statistical dependencies in MLMs and show that these dependencies encode useful inductive biases in the form of syntactic structures. In an unsupervised parsing evaluation, simply forming a minimum spanning tree on the implied statistical dependence structure outperforms a classic method for unsupervised parsing (58.74 vs. 55.91 UUAS).
We consider a discounted infinite horizon optimal stopping problem. If the underlying distribution is known a priori, the solution of this problem is obtained via dynamic programming (DP) and is given by a well known threshold rule. When information
on this distribution is lacking, a natural (though naive) approach is explore-then-exploit, whereby the unknown distribution or its parameters are estimated over an initial exploration phase, and this estimate is then used in the DP to determine actions over the residual exploitation phase. We show: (i) with proper tuning, this approach leads to performance comparable to the full information DP solution; and (ii) despite common wisdom on the sensitivity of such plug in approaches in DP due to propagation of estimation errors, a surprisingly short (logarithmic in the horizon) exploration horizon suffices to obtain said performance. In cases where the underlying distribution is heavy-tailed, these observations are even more pronounced: a ${it single , sample}$ exploration phase suffices.
Semiconducting Transition Metal Dichalcogenides (TMDs) have significant nonlinear optical effects. In this work we have used second-harmonic generation (SHG) and the four-wave mixing (FWM) spectroscopy in resonance with the excitons in MoS2, MoSe2, a
nd WS2 monolayers to characterize the nonlinear optical properties of these materials. We show that trions and excitons are responsible for enhancing the nonlinear optical response, and determine the exciton and trion energies by comparing with the photoluminescence spectra. Moreover, we extract the second and third order optical sheet susceptibility near exciton energies and compare with values found in the literature. We also demonstrate the ability to generate different nonlinear effects in a wide spectral range in the visible region for monolayer MoS2, opening the possibility of using two-dimensional materials for nonlinear optoelectronic and photonic applications.
Radio access network (RAN) virtualization is gaining more and more ground and expected to re-architect the next-generation cellular networks. Existing RAN virtualization studies and solutions have mostly focused on sharing communication capacity and
tend to require the use of the same PHY and MAC layers across network slices. This approach has not considered the scenarios where different slices require different PHY and MAC layers, for instance, for radically different services and for whole-stack research in wireless living labs where novel PHY and MAC layers need to be deployed concurrently with existing ones on the same physical infrastructure. To enable whole-stack slicing where different PHY and MAC layers may be deployed in different slices, we develop PV-RAN, the first open-source virtual RAN platform that enables the sharing of the same SDR physical resources across multiple slices. Through API Remoting, PV-RAN enables running paravirtualized instances of OpenAirInterface (OAI) at different slices without requiring modifying OAI source code. PV-RAN effectively leverages the inter-domain communication mechanisms of Xen to transport time-sensitive I/Q samples via shared memory, making the virtualization overhead in communication almost negligible. We conduct detailed performance benchmarking of PV-RAN and demonstrate its low overhead and high efficiency. We also integrate PV-RAN with the CyNet wireless living lab for smart agriculture and transportation.
This paper is a study of fine-tuning of BERT contextual representations, with focus on commonly observed instabilities in few-sample scenarios. We identify several factors that cause this instability: the common use of a non-standard optimization met
hod with biased gradient estimation; the limited applicability of significant parts of the BERT network for down-stream tasks; and the prevalent practice of using a pre-determined, and small number of training iterations. We empirically test the impact of these factors, and identify alternative practices that resolve the commonly observed instability of the process. In light of these observations, we re-visit recently proposed methods to improve few-sample fine-tuning with BERT and re-evaluate their effectiveness. Generally, we observe the impact of these methods diminishes significantly with our modified process.
Single-layer heterostructures exhibit striking quasiparticle properties and many-body interaction effects that hold promise for a range of applications. However, their properties can be altered by intrinsic and extrinsic defects, thus diminishing the
ir applicability. Therefore, it is of paramount importance to identify defects and understand 2D materials degradation over time using advanced multimodal imaging techniques as well as stabilize degradation via built-in interface protection. Here we implemented a liquid-phase precursor approach to synthesize 2D in-plane MoS2-WS2 heterostructures exhibiting nanoscale alloyed interfaces and map exotic interface effects during photo-degradation using a novel combination of hyperspectral tip-enhanced photoluminescence, Raman and near-field nanoscopy. Surprisingly, 2D alloyed regions exhibit remarkable thermal and photo-degradation stability providing protection against oxidation. Coupled with surface and interface strain, 2D alloy regions create localized potential wells that concentrate excitonic species via a charge carrier funneling effect. These results provide a clear understanding of the importance of 2D alloys as systems able to withstand degradation effects over time, and could be now used to stabilize optoelectronic devices based on 2D materials.
Not all data in a typical training set help with generalization; some samples can be overly ambiguous or outrightly mislabeled. This paper introduces a new method to identify such samples and mitigate their impact when training neural networks. At th
e heart of our algorithm is the Area Under the Margin (AUM) statistic, which exploits differences in the training dynamics of clean and mislabeled samples. A simple procedure - adding an extra class populated with purposefully mislabeled threshold samples - learns a AUM upper bound that isolates mislabeled data. This approach consistently improves upon prior work on synthetic and real-world datasets. On the WebVision50 classification task our method removes 17% of training data, yielding a 1.6% (absolute) improvement in test error. On CIFAR100 removing 13% of the data leads to a 1.2% drop in error.
