Search

Most active learning research has focused on methods which perform well when many labels are available, but can be dramatically worse than random selection when label budgets are small. Other methods have focused on the low-budget regime, but do poorly as label budgets increase. As the line between "low" and "high" budgets varies by problem, this is a serious issue in practice. We propose uncertainty coverage, an objective which generalizes a variety of low- and high-budget objectives, as well as natural, hyperparameter-light methods to smoothly interpolate between low- and high-budget regimes. We call greedy optimization of the estimate Uncertainty Herding; this simple method is computationally fast, and we prove that it nearly optimizes the distribution-level coverage. In experimental validation across a variety of active learning tasks, our proposal matches or beats state-of-the-art performance in essentially all cases; it is the only method of which we are aware that reliably works well in both low- and high-budget settings.

Time series~(TS) modeling is essential in dynamic systems like weather prediction and anomaly detection. Recent studies utilize Large Language Models (LLMs) for TS modeling, leveraging their powerful pattern recognition capabilities. These methods primarily position LLMs as the predictive backbone, often omitting the mathematical modeling within traditional TS models, such as periodicity. However, disregarding the potential of LLMs also overlooks their pattern recognition capabilities. To address this gap, we introduce \textit{LLM-TS Integrator}, a novel framework that effectively integrates the capabilities of LLMs into traditional TS modeling. Central to this integration is our \textit{mutual information} module. The core of this \textit{mutual information} module is a traditional TS model enhanced with LLM-derived insights for improved predictive abilities. This enhancement is achieved by maximizing the mutual information between traditional model's TS representations and LLM's textual representation counterparts, bridging the two modalities. Moreover, we recognize that samples vary in importance for two losses: traditional prediction and mutual information maximization. To address this variability, we introduce the \textit{sample reweighting} module to improve information utilization. This module assigns dual weights to each sample: one for prediction loss and another for mutual information loss, dynamically optimizing these weights via bi-level optimization. Our method achieves state-of-the-art or comparable performance across five mainstream TS tasks, including short-term and long-term forecasting, imputation, classification, and anomaly detection., Comment: 18 pages, 13 figures, 18 tables

We use James Webb Space Telescope (JWST) Mid-Infrared Instrument (MIRI) medium-resolution spectrometer (MRS) observations of 3C 293 (UGC 8782), CGCG 012-070 and NGC 3884 to investigate the origin of the H$_2$ emission. These three nearby Active Galactic Nucleus (AGN) hosts are known to present H$_2$ emission excess relative to star-forming galaxies, as traced by the H$_2$ S(3)/PAH$_{\rm 11.3\mu m}$ line ratio. We define the kinematically disturbed region (KDR) by the AGN and the virially dominated region (VDR) based on the H$_2$ line widths, using the $W{\rm 80}$ parameter. From the correlations between $W{\rm 80}$ and H$_2$ S(3)/PAH${\rm 11.3\mu m}$, as well as the higher H$2$ S(5)/H$2$ S(3) and [Fe II]${\rm 5.34 \mu m}$/PAH${\rm 11.3\mu m}$ ratios and flatter power-law temperature distributions observed in the KDR, we conclude that the H$_2$ emission in the KDR is primarily driven by shock-heated gas. For 3C 293, the KDR is co-spatial with the radio core, indicating that the origin of the shocks is the interaction of the radio jet with the interstellar medium, which is also responsible for the observed molecular and ionized gas outflows in this source. The other galaxies are weak radio sources; however, due to the lack of high-resolution radio images, we cannot rule out low-power jets as the origin of the shock-heated H$_2$. Our results indicate that the excess H$_2$ emission excess is associated to shock heating of the gas, generated by outflows or by the interaction of the radio jet with the ambient gas., Comment: 15 pages, 5 figures, submitted to ApJ. Updated version

We use the James Webb Space Telescope (JWST) Mid-Infrared Instrument (MIRI) medium-resolution spectrometer (MRS) observations of the radio loud AGN host UGC 8782 to map the warm molecular and ionized gas kinematics. The data reveal spatially resolved outflows in the inner 2 kpc, seen in low ionization (traced by the [Ar ii] 6.99um emission) and in warm molecular gas (traced by the H2 rotational transitions). We find a maximum mass-outflow rate of 4.90 (2.04) M/yr at ~900 pc from the nucleus for the warm outflow (198 K< T< 1000 K) and estimate and outflow rate of up to 1.22 (0.51) M/yr for the hotter gas phase (T > 1000 K). These outflows can clear the entire nuclear reservoir of warm molecular gas in about 1 Myr. The derived kinetic power of the molecular outflows lead to coupling efficiencies of 2-5 percent of the AGN luminosity, way above the minimum expected to the AGN feedback be effective quenching the star formation., Comment: Accepted for publication at APJ

Despite attaining high empirical generalization, the sharpness of models trained with sharpness-aware minimization (SAM) do not always correlate with generalization error. Instead of viewing SAM as minimizing sharpness to improve generalization, our paper considers a new perspective based on SAM's training dynamics. We propose that perturbations in SAM perform perturbed forgetting, where they discard undesirable model biases to exhibit learning signals that generalize better. We relate our notion of forgetting to the information bottleneck principle, use it to explain observations like the better generalization of smaller perturbation batches, and show that perturbed forgetting can exhibit a stronger correlation with generalization than flatness. While standard SAM targets model biases exposed by the steepest ascent directions, we propose a new perturbation that targets biases exposed through the model's outputs. Our output bias forgetting perturbations outperform standard SAM, GSAM, and ASAM on ImageNet, robustness benchmarks, and transfer to CIFAR-{10,100}, while sometimes converging to sharper regions. Our results suggest that the benefits of SAM can be explained by alternative mechanistic principles that do not require flatness of the loss surface., Comment: Published as a conference paper at ICML 2024. 9 pages main, 15 pages total including references and appendix

Although neural networks are conventionally optimized towards zero training loss, it has been recently learned that targeting a non-zero training loss threshold, referred to as a flood level, often enables better test time generalization. Current approaches, however, apply the same constant flood level to all training samples, which inherently assumes all the samples have the same difficulty. We present AdaFlood, a novel flood regularization method that adapts the flood level of each training sample according to the difficulty of the sample. Intuitively, since training samples are not equal in difficulty, the target training loss should be conditioned on the instance. Experiments on datasets covering four diverse input modalities - text, images, asynchronous event sequences, and tabular - demonstrate the versatility of AdaFlood across data domains and noise levels.

Multi-task networks rely on effective parameter sharing to achieve robust generalization across tasks. In this paper, we present a novel parameter sharing method for multi-task learning that conditions parameter sharing on both the task and the intermediate feature representations at inference time. In contrast to traditional parameter sharing approaches, which fix or learn a deterministic sharing pattern during training and apply the same pattern to all examples during inference, we propose to dynamically decide which parts of the network to activate based on both the task and the input instance. Our approach learns a hierarchical gating policy consisting of a task-specific policy for coarse layer selection and gating units for individual input instances, which work together to determine the execution path at inference time. Experiments on the NYU v2, Cityscapes and MIMIC-III datasets demonstrate the potential of the proposed approach and its applicability across problem domains.

We use optical Integral Field Spectroscopy (IFU) to study the gas emission structure and kinematics in the inner 3.4$\times$4.9 kpc$^2$ region of the galaxy UGC 8782 (3C 293), host of a radio loud Active Galactic Nucleus (AGN). The observations were performed with the GMOS-IFU on the Gemini North telescope, resulting in a spatial resolution of $\sim725$ pc at the distance of the galaxy. While the stars present ordered rotation following the orientation of the large scale disc, the gas shows a disturbed kinematics. The emission-line profiles present two kinematic components: a narrow ($\sigma\lesssim200$ km s$^{-1}$) component associated with the gas in the disc of the galaxy and a broad ($\sigma\gtrsim200$ km s$^{-1}$) component produced by gas outflows. Emission-line ratio diagrams indicate that the gas in the disc is excited by the AGN radiation field, while the emission of the outflow includes additional contribution of shock excitation due to the interaction of the radio jet with the environment gas. Deviations from pure rotation, of up to 30 km s$^{-1}$, are observed in the disc component and likely produced by a previous merger event. The broad component is blueshifted by $\sim150-500$ km s$^{-1}$ relative to the systemic velocity of the galaxy in all locations. We construct radial profiles of the mass outflow rate and kinetic power of the ionized gas outflows, which have the maximum values at $\sim1$ kpc from the nucleus with peak values of $\dot{M}_{\rm out,\Delta R}=0.5\pm0.1$ M$_\odot$ yr$^{-1}$ and $\dot{K}_{\rm out,\Delta R} =$(6.8$\pm$1.1)$\times$10$^{41}$ erg s$^{-1}$. The kinetic coupling efficiency of these outflows is in the range of 1$-$3 per cent, indicating that they could be powerful enough to affect the star formation in the host galaxy as predicted by theoretical simulations., Comment: Accepted for publication by MNRAS

A temporal point process (TPP) is a stochastic process where its realization is a sequence of discrete events in time. Recent work in TPPs model the process using a neural network in a supervised learning framework, where a training set is a collection of all the sequences. In this work, we propose to train TPPs in a meta learning framework, where each sequence is treated as a different task, via a novel framing of TPPs as neural processes (NPs). We introduce context sets to model TPPs as an instantiation of NPs. Motivated by attentive NP, we also introduce local history matching to help learn more informative features. We demonstrate the potential of the proposed method on popular public benchmark datasets and tasks, and compare with state-of-the-art TPP methods., Comment: Accepted to ICLR2023

ML models often operate within the context of a larger system that can adapt its response when the ML model is uncertain, such as falling back on safe defaults or a human in the loop. This commonly encountered operational context calls for principled techniques for training ML models with the option to abstain from predicting when uncertain. Selective neural networks are trained with an integrated option to abstain, allowing them to learn to recognize and optimize for the subset of the data distribution for which confident predictions can be made. However, optimizing selective networks is challenging due to the non-differentiability of the binary selection function (the discrete decision of whether to predict or abstain). This paper presents a general method for training selective networks that leverages the Gumbel-softmax reparameterization trick to enable selection within an end-to-end differentiable training framework. Experiments on public datasets demonstrate the potential of Gumbel-softmax selective networks for selective regression and classification.

This paper documents the seventeenth data release (DR17) from the Sloan Digital Sky Surveys; the fifth and final release from the fourth phase (SDSS-IV). DR17 contains the complete release of the Mapping Nearby Galaxies at Apache Point Observatory (MaNGA) survey, which reached its goal of surveying over 10,000 nearby galaxies. The complete release of the MaNGA Stellar Library (MaStar) accompanies this data, providing observations of almost 30,000 stars through the MaNGA instrument during bright time. DR17 also contains the complete release of the Apache Point Observatory Galactic Evolution Experiment 2 (APOGEE-2) survey which publicly releases infra-red spectra of over 650,000 stars. The main sample from the Extended Baryon Oscillation Spectroscopic Survey (eBOSS), as well as the sub-survey Time Domain Spectroscopic Survey (TDSS) data were fully released in DR16. New single-fiber optical spectroscopy released in DR17 is from the SPectroscipic IDentification of ERosita Survey (SPIDERS) sub-survey and the eBOSS-RM program. Along with the primary data sets, DR17 includes 25 new or updated Value Added Catalogs (VACs). This paper concludes the release of SDSS-IV survey data. SDSS continues into its fifth phase with observations already underway for the Milky Way Mapper (MWM), Local Volume Mapper (LVM) and Black Hole Mapper (BHM) surveys., Comment: 40 pages, 8 figures, 6 tables. In press at ApJSS (arxiv v2 corrects some minor typos and updates references)

Predicting multiple heterogeneous biological and medical targets is a challenge for traditional deep learning models. In contrast to single-task learning, in which a separate model is trained for each target, multi-task learning (MTL) optimizes a single model to predict multiple related targets simultaneously. To address this challenge, we propose the Multi-gate Mixture-of-Experts with Exclusivity (MMoEEx). Our work aims to tackle the heterogeneous MTL setting, in which the same model optimizes multiple tasks with different characteristics. Such a scenario can overwhelm current MTL approaches due to the challenges in balancing shared and task-specific representations and the need to optimize tasks with competing optimization paths. Our method makes two key contributions: first, we introduce an approach to induce more diversity among experts, thus creating representations more suitable for highly imbalanced and heterogenous MTL learning; second, we adopt a two-step optimization [6, 11] approach to balancing the tasks at the gradient level. We validate our method on three MTL benchmark datasets, including Medical Information Mart for Intensive Care (MIMIC-III) and PubChem BioAssay (PCBA)., Comment: 10 pages, 7 figures, BIOKDD, IEEE/ACM

Single encoder-decoder methodologies for semantic segmentation are reaching their peak in terms of segmentation quality and efficiency per number of layers. To address these limitations, we propose a new architecture based on a decoder which uses a set of shallow networks for capturing more information content. The new decoder has a new topology of skip connections, namely backward and stacked residual connections. In order to further improve the architecture we introduce a weight function which aims to re-balance classes to increase the attention of the networks to under-represented objects. We carried out an extensive set of experiments that yielded state-of-the-art results for the CamVid, Gatech and Freiburg Forest datasets. Moreover, to further prove the effectiveness of our decoder, we conducted a set of experiments studying the impact of our decoder to state-of-the-art segmentation techniques. Additionally, we present a set of experiments augmenting semantic segmentation with optical flow information, showing that motion clues can boost pure image based semantic segmentation approaches.

This paper presents a calibration method for the speed-flow model of the HCM 2016 for freeways and multilane highways allied to temporal analysis of traffic stream. The proposed method was developed using a sample of more than one million observations collected by 23 traffic sensors on four highways in the state of S{\~a}o Paulo. The method is structured on Bayesian inference and provided for each model parameters a probability distribution function. The free-flow speed and capacity presented a probability density function that approximates a Normal distribution. The segment in which the speed of traffic stream remain constant with the increase of the traffic flow is lower than described in HCM 2016, being in some cases close to zero. Along with the proposed calibration method an analysis of temporal variation is performed which shows a significant variation in traffic behavior for different periods. The free-flow speed, capacity and breakpoint distributions obtained through monthly and annual calibration were considered equal by means of Kolmogorov-Smirnov test, different for the model calibration coefficient., Comment: 13 pages; 43 figures

This paper presents an experimental study on the transition from smooth to rough walls, and back to the smooth one, in turbulent closed-conduit flows. These transitions cause a shift on flow velocity profiles that changes their parameters when compared to the flow over a smooth wall. Different water flow rates were imposed in a closed conduit of rectangular cross section, where rough elements consisting of cavities of d- and k-type were positioned covering a part of the bottom wall of the test section. Reynolds numbers based on the channel half-height were moderate, varying between 7800 and 9600, and the regime upstream of the rough elements was hydraulically smooth. Experimental data for this specific case remain scarce and the involved physics rests to be understood. The flow field was measured by low frequency PIV (particle image velocimetry) and by flow visualization, the latter using a continuous 0.1 W laser, a high-speed camera, and scripts written by the authors. From the instantaneous fields measured with PIV, the mean velocities, fluctuations, shear stresses and turbulence production were computed. The results show the presence of oscillations in Reynolds stress and turbulence production, that are higher for the k-type roughness and were not shown in previous experimental works. From the high-speed movies, the angular velocities and frequencies of vortices in the cavities were computed, and the occurrence of fluid ejection from the cavities to upper layers of the flow was observed. A relation between the angular velocities of inner-cavities vortices and the oscillations in Reynolds stress and turbulence production is proposed., Comment: This is a pre-print of an article published in Journal of the Brazilian Society of Mechanical Sciences and Engineering, 40:187, 2018. The final authenticated version is available online at: http://dx.doi.org/10.1007/s40430-018-1113-9

This paper presents an experimental investigation of the perturbation of a turbulent closed-conduit flow by two-dimensional triangular ripples. Two ripple configurations were employed: one single asymmetric triangular ripple, and two consecutive asymmetric triangular ripples, all of them with the same geometry. Different water flows were imposed over either one or two ripples fixed to the bottom wall of a closed conduit, and the flow field was measured by PIV (particle image velocimetry). Reynolds numbers based on the channel height were moderate, varying between 27500 and 34700. The regime was hydraulically smooth, and the blockage ratio was significant. Experimental data for this specific case remain scarce, and the physics involved has yet to be fully elucidated. Using the instantaneous flow fields, the mean velocities and fluctuations were computed, and the shear stresses over the ripples were determined. The velocities and stresses obtained in this way for the single ripple and for the pair of ripples are compared, and the surface shear stress is discussed in terms of bed stability. Our results show that for the single and upstream ripples the shear stress increases at the leading edge and decreases toward the crest, while for the downstream ripple it decreases monotonically between the reattachment point and the crest. The stress distribution over the downstream ripple, which is different from both the upstream and single ripples, is shown to sustain existing ripples over loose beds., Comment: This manuscript version is made available under the CC-BY-NC-ND 4.0 license https://creativecommons.org/licenses/by-nc-nd/4.0/. This is a pre-print of an article published in Journal of the Brazilian Society of Mechanical Sciences and Engineering, 40:138, 2018. The final authenticated version is available online at: http://dx.doi.org/10.1007/s40430-018-1050-7

Compared to LiDAR-based localization methods, which provide high accuracy but rely on expensive sensors, visual localization approaches only require a camera and thus are more cost-effective while their accuracy and reliability typically is inferior to LiDAR-based methods. In this work, we propose a vision-based localization approach that learns from LiDAR-based localization methods by using their output as training data, thus combining a cheap, passive sensor with an accuracy that is on-par with LiDAR-based localization. The approach consists of two deep networks trained on visual odometry and topological localization, respectively, and a successive optimization to combine the predictions of these two networks. We evaluate the approach on a new challenging pedestrian-based dataset captured over the course of six months in varying weather conditions with a high degree of noise. The experiments demonstrate that the localization errors are up to 10 times smaller than with traditional vision-based localization methods., Comment: 16 pages, 7 figures, ISRR 2017 submission

Robots are expected to operate autonomously in dynamic environments. Understanding the underlying dynamic characteristics of objects is a key enabler for achieving this goal. In this paper, we propose a method for pointwise semantic classification of 3D LiDAR data into three classes: non-movable, movable and dynamic. We concentrate on understanding these specific semantics because they characterize important information required for an autonomous system. Non-movable points in the scene belong to unchanging segments of the environment, whereas the remaining classes corresponds to the changing parts of the scene. The difference between the movable and dynamic class is their motion state. The dynamic points can be perceived as moving, whereas movable objects can move, but are perceived as static. To learn the distinction between movable and non-movable points in the environment, we introduce an approach based on deep neural network and for detecting the dynamic points, we estimate pointwise motion. We propose a Bayes filter framework for combining the learned semantic cues with the motion cues to infer the required semantic classification. In extensive experiments, we compare our approach with other methods on a standard benchmark dataset and report competitive results in comparison to the existing state-of-the-art. Furthermore, we show an improvement in the classification of points by combining the semantic cues retrieved from the neural network with the motion cues., Comment: 8 pages to be published in IROS 2017

General human action recognition requires understanding of various visual cues. In this paper, we propose a network architecture that computes and integrates the most important visual cues for action recognition: pose, motion, and the raw images. For the integration, we introduce a Markov chain model which adds cues successively. The resulting approach is efficient and applicable to action classification as well as to spatial and temporal action localization. The two contributions clearly improve the performance over respective baselines. The overall approach achieves state-of-the-art action classification performance on HMDB51, J-HMDB and NTU RGB+D datasets. Moreover, it yields state-of-the-art spatio-temporal action localization results on UCF101 and J-HMDB., Comment: 10 pages, 7 figures, ICCV 2017 submission