Search

To interact with humans and act in the world, agents need to understand the range of language that people use and relate it to the visual world. While current agents can learn to execute simple language instructions, we aim to build agents that leverage diverse language -- language like "this button turns on the TV" or "I put the bowls away" -- that conveys general knowledge, describes the state of the world, provides interactive feedback, and more. Our key idea is that agents should interpret such diverse language as a signal that helps them predict the future: what they will observe, how the world will behave, and which situations will be rewarded. This perspective unifies language understanding with future prediction as a powerful self-supervised learning objective. We instantiate this in Dynalang, an agent that learns a multimodal world model to predict future text and image representations, and learns to act from imagined model rollouts. While current methods that learn language-conditioned policies degrade in performance with more diverse types of language, we show that Dynalang learns to leverage environment descriptions, game rules, and instructions to excel on tasks ranging from game-playing to navigating photorealistic home scans. Finally, we show that our method enables additional capabilities due to learning a generative model: Dynalang can be pretrained on text-only data, enabling learning from offline datasets, and generate language grounded in an environment., Comment: ICML 2024. Website: https://dynalang.github.io/

Artificial intelligence (AI) reasoning and explainable AI (XAI) tasks have gained popularity recently, enabling users to explain the predictions or decision processes of AI models. This paper introduces Forest Monkey (FM), a toolkit designed to reason the outputs of any AI-based defect detection and/or classification model with data explainability. Implemented as a Python package, FM takes input in the form of dataset folder paths (including original images, ground truth labels, and predicted labels) and provides a set of charts and a text file to illustrate the reasoning results and suggest possible improvements. The FM toolkit consists of processes such as feature extraction from predictions to reasoning targets, feature extraction from images to defect characteristics, and a decision tree-based AI-Reasoner. Additionally, this paper investigates the time performance of the FM toolkit when applied to four AI models with different datasets. Lastly, a tutorial is provided to guide users in performing reasoning tasks using the FM toolkit., Comment: 6 pages, 5 figures, accepted in 2023 IEEE symposium series on computational intelligence (SSCI)

Neuromorphic processors have garnered considerable interest in recent years for their potential in energy-efficient and high-speed computing. The Locally Competitive Algorithm (LCA) has been utilized for power efficient sparse coding on neuromorphic processors, including the first Loihi processor. With the Loihi 2 processor enabling custom neuron models and graded spike communication, more complex implementations of LCA are possible. We present a new implementation of LCA designed for the Loihi 2 processor and perform an initial set of benchmarks comparing it to LCA on CPU and GPU devices. In these experiments LCA on Loihi 2 is orders of magnitude more efficient and faster for large sparsity penalties, while maintaining similar reconstruction quality. We find this performance improvement increases as the LCA parameters are tuned towards greater representation sparsity. Our study highlights the potential of neuromorphic processors, particularly Loihi 2, in enabling intelligent, autonomous, real-time processing on small robots, satellites where there are strict SWaP (small, lightweight, and low power) requirements. By demonstrating the superior performance of LCA on Loihi 2 compared to conventional computing device, our study suggests that Loihi 2 could be a valuable tool in advancing these types of applications. Overall, our study highlights the potential of neuromorphic processors for efficient and accurate data processing on resource-constrained devices.

In machine learning for sequential decision-making, an algorithmic agent learns to interact with an environment while receiving feedback in the form of a reward signal. However, in many unstructured real-world settings, such a reward signal is unknown and humans cannot reliably craft a reward signal that correctly captures desired behavior. To solve tasks in such unstructured and open-ended environments, we present Demonstration-Inferred Preference Reinforcement Learning (DIP-RL), an algorithm that leverages human demonstrations in three distinct ways, including training an autoencoder, seeding reinforcement learning (RL) training batches with demonstration data, and inferring preferences over behaviors to learn a reward function to guide RL. We evaluate DIP-RL in a tree-chopping task in Minecraft. Results suggest that the method can guide an RL agent to learn a reward function that reflects human preferences and that DIP-RL performs competitively relative to baselines. DIP-RL is inspired by our previous work on combining demonstrations and pairwise preferences in Minecraft, which was awarded a research prize at the 2022 NeurIPS MineRL BASALT competition, Learning from Human Feedback in Minecraft. Example trajectory rollouts of DIP-RL and baselines are located at https://sites.google.com/view/dip-rl., Comment: Paper accepted at The Many Facets of Preference Learning Workshop at the International Conference on Machine Learning (ICML), Honolulu, Hawaii, USA, 2023

The SPICE/HeRALD collaboration is performing R&D to enable studies of sub-GeV dark matter models using a variety of target materials. Here we report our recent progress on instrumenting a superfluid $^4$He target mass with a transition-edge sensor based calorimeter to detect both atomic signals (scintillation) and $^4$He quasiparticle (phonon and roton) excitations. The sensitivity of HeRALD to the critical "quantum evaporation" signal from $^4$He quasiparticles requires us to block the superfluid film flow to the calorimeter. We have developed a heat-free film-blocking method employing an unoxidized Cs film, which we implemented in a prototype "HeRALD v0.1" detector of ~10 g target mass. This article reports initial studies of the atomic and quasiparticle signal channels. A key result of this work is the measurement of the quantum evaporation channel's gain of 0.15 $\pm$ 0.01, which will enable $^4$He-based dark matter experiments in the near term. With this gain the HeRALD detector reported here has an energy threshold of 145 eV at 5 sigma, which would be sensitive to dark matter masses down to 220 MeV/c$^2$., Comment: 14 pages, 9 figures

As the use of artificial intelligent (AI) models becomes more prevalent in industries such as engineering and manufacturing, it is essential that these models provide transparent reasoning behind their predictions. This paper proposes the AI-Reasoner, which extracts the morphological characteristics of defects (DefChars) from images and utilises decision trees to reason with the DefChar values. Thereafter, the AI-Reasoner exports visualisations (i.e. charts) and textual explanations to provide insights into outputs made by masked-based defect detection and classification models. It also provides effective mitigation strategies to enhance data pre-processing and overall model performance. The AI-Reasoner was tested on explaining the outputs of an IE Mask R-CNN model using a set of 366 images containing defects. The results demonstrated its effectiveness in explaining the IE Mask R-CNN model's predictions. Overall, the proposed AI-Reasoner provides a solution for improving the performance of AI models in industrial applications that require defect analysis., Comment: 8 pages, 3 figures, 5 tables; accepted in 2023 IEEE symposium series on computational intelligence (SSCI)

We investigate the ionised gas morphology, excitation properties, and kinematics in 19 nearby star-forming galaxies from the PHANGS-MUSE survey. We directly compare the kinetic energy of expanding superbubbles and the turbulent motions in the interstellar medium with the mechanical energy deposited by massive stars in the form of winds and supernovae, with the aim to answer whether the stellar feedback is responsible for the observed turbulent motions and to quantify the fraction of mechanical energy retained in the superbubbles. Based on the distribution of the flux and velocity dispersion in the H$\alpha$ line, we select 1484 regions of locally elevated velocity dispersion ($\sigma$(H$\alpha$)>45 km/s), including at least 171 expanding superbubbles. We analyse these regions and relate their properties to those of the young stellar associations and star clusters identified in PHANGS-HST data. We find a good correlation between the kinetic energy of the ionised gas and the total mechanical energy input from supernovae and stellar winds from the stellar associations, with a typical coupling efficiency of 10-20%. The contribution of mechanical energy by the supernovae alone is not sufficient to explain the measured kinetic energy of the ionised gas, which implies that pre-supernova feedback in the form of radiation/thermal pressure and winds is necessary. We find that the gas kinetic energy decreases with metallicity for our sample covering Z=0.5-1.0 Zsun, reflecting the lower impact of stellar feedback. For the sample of superbubbles, we find that about 40% of the young stellar associations are preferentially located in their rims. We also find a slightly higher (by ~15%) fraction of the youngest (<3 Myr) stellar associations in the rims of the superbubbles than in the centres, and the opposite for older associations, which implies possible propagation or triggering of star formation., Comment: Accepted by A&A. The abstract is abridged. 31 pages (including 8 pages in appendix), 20 figures, 2 tables

Measurements of the star formation activity on cloud scales are fundamental to uncovering the physics of the molecular cloud, star formation, and stellar feedback cycle in galaxies. Infrared (IR) emission from small dust grains and polycyclic aromatic hydrocarbons (PAHs) are widely used to trace the obscured component of star formation. However, the relation between these emission features and dust attenuation is complicated by the combined effects of dust heating from old stellar populations and an uncertain dust geometry with respect to heating sources. We use images obtained with NIRCam and MIRI as part of the PHANGS--JWST survey to calibrate dust emission at 21$\rm \mu m$, and the emission in the PAH-tracing bands at 3.3, 7.7, 10, and 11.3$\rm \mu m$ as tracers of obscured star formation. We analyse $\sim$ 20000 optically selected HII regions across 19 nearby star-forming galaxies, and benchmark their IR emission against dust attenuation measured from the Balmer decrement. We model the extinction-corrected H$\alpha$ flux as the sum of the observed H$\alpha$ emission and a term proportional to the IR emission, with $a_{IR}$ as the proportionality coefficient. A constant $a_{IR}$ leads to extinction-corrected H$\alpha$ estimates which agree with those obtained with the Balmer decrement with a scatter of $\sim$ 0.1 dex for all bands considered. Among these bands, 21$\rm \mu m$ emission is demonstrated to be the best tracer of dust attenuation. The PAH-tracing bands underestimate the correction for bright HII regions, since in these environments the ratio of PAH-tracing bands to 21$\rm \mu m$ decreases, signalling destruction of the PAH molecules. For fainter HII regions all bands suffer from an increasing contamination from the diffuse infrared background., Comment: accepted for publication in A&A

Structure-based molecular ML (SBML) models can be highly sensitive to input geometries and give predictions with large variance. We present an approach to mitigate the challenge of selecting conformations for such models by generating conformers that explicitly minimize predictive uncertainty. To achieve this, we compute estimates of aleatoric and epistemic uncertainties that are differentiable w.r.t. latent posteriors. We then iteratively sample new latents in the direction of lower uncertainty by gradient descent. As we train our predictive models jointly with a conformer decoder, the new latent embeddings can be mapped to their corresponding inputs, which we call \textit{MoleCLUEs}, or (molecular) counterfactual latent uncertainty explanations \citep{antoran2020getting}. We assess our algorithm for the task of predicting drug properties from 3D structure with maximum confidence. We additionally analyze the structure trajectories obtained from conformer optimizations, which provide insight into the sources of uncertainty in SBML., Comment: NeurIPS 2023 AI for Science Workshop

One of the last remaining frontiers in optical/near-infrared observational astronomy is the low surface brightness regime (LSB, V-band surface brightness, $\mu_V>$ 27 AB mag/arcsec$^2$). These are the structures at very low stellar surface densities, largely unseen by even current wide-field surveys such as the Legacy Survey. Studying this domain promises to be transformative for our understanding of star formation in low-mass galaxies, the hierarchical assembly of galaxies and galaxy clusters, and the nature of dark matter. It is thus essential to reach depths beyond $\mu_V$ = 30 AB mag/arcsec$^2$ to detect the faintest extragalactic sources, such as dwarf galaxies and the stellar halos around galaxies and within galaxy clusters. The High Latitude Wide Area Survey offers a unique opportunity to statistically study the LSB universe at unprecedented depths in the IR over an area of $\sim$2000 square degrees. The high spatial resolution will minimize source confusion, allowing an unbiased characterization of LSB structures, including the identification of stars in nearby LSB galaxies and globular clusters. In addition, the combination of Roman with other upcoming deep imaging observatories (such as Rubin) will provide multi-wavelength coverage to derive photometric redshifts and infer the stellar populations of LSB objects., Comment: White paper submitted to the call for input for the Roman Space Telescope's Core Community Surveys

We propose a new score-based approach to generate 3D molecules represented as atomic densities on regular grids. First, we train a denoising neural network that learns to map from a smooth distribution of noisy molecules to the distribution of real molecules. Then, we follow the neural empirical Bayes framework (Saremi and Hyvarinen, 19) and generate molecules in two steps: (i) sample noisy density grids from a smooth distribution via underdamped Langevin Markov chain Monte Carlo, and (ii) recover the "clean" molecule by denoising the noisy grid with a single step. Our method, VoxMol, generates molecules in a fundamentally different way than the current state of the art (ie, diffusion models applied to atom point clouds). It differs in terms of the data representation, the noise model, the network architecture and the generative modeling algorithm. Our experiments show that VoxMol captures the distribution of drug-like molecules better than state of the art, while being faster to generate samples.

$^{180m}$Ta is a rare nuclear isomer whose decay has never been observed. Its remarkably long lifetime surpasses the half-lives of all other known $\beta$ and electron capture decays due to the large K-spin differences and small energy differences between the isomeric and lower energy states. Detecting its decay presents a significant experimental challenge but could shed light on neutrino-induced nucleosynthesis mechanisms, the nature of dark matter and K-spin violation. For this study, we repurposed the MAJORANA DEMONSTRATOR, an experimental search for the neutrinoless double-beta decay of $^{76}$Ge using an array of high-purity germanium detectors, to search for the decay of $^{180m}$Ta. More than 17 kilograms, the largest amount of tantalum metal ever used for such a search was installed within the ultra-low background detector array. In this paper we present results from the first year of Ta data taking and provide an updated limit for the $^{180m}$Ta half-life on the different decay channels. With new limits up to 1.5 x $10^{19}$ years, we improved existing limits by one to two orders of magnitude. This result is the most sensitive search for a single $\beta$ and electron capture decay ever achieved.

The morphology of a galaxy stems from secular and environmental processes during its evolutionary history. Thus galaxy morphologies have been a long used tool to gain insights on galaxy evolution. We visually classify morphologies on cloud-scales based on the molecular gas distribution of a large sample of 79 nearby main-sequence galaxies, using 1'' resolution CO(2-1) ALMA observations taken as part of the PHANGS survey. To do so, we devise a morphology classification scheme for different types of bars, spiral arms (grand-design, flocculent, multi-arm and smooth), rings (central and non-central rings) similar to the well-established optical ones, and further introduce bar lane classes. In general, our cold gas based morphologies agree well with the ones based on stellar light. Both our bars as well as grand-design spiral arms are preferentially found at the higher mass end of our sample. Our gas-based classification indicates a potential for misidentification of unbarred galaxies in the optical when massive star formation is present. Central or nuclear rings are present in a third of the sample with a strong preferences for barred galaxies (59%). As stellar bars are present in 45$\pm$5% of our sample galaxies, we explore the utility of molecular gas as tracer of bar lane properties. We find that more curved bar lanes have a shorter radial extent in molecular gas and reside in galaxies with lower molecular to stellar mass ratios than those with straighter geometries. Galaxies display a wide range of CO morphology, and this work provides a catalogue of morphological features in a representative sample of nearby galaxies., Comment: 17 pages, 14 figures (+ Appendix 9 pages, 4 figures). Accepted for publication in A&A

Learning from human feedback has been shown to improve text-to-image models. These techniques first learn a reward function that captures what humans care about in the task and then improve the models based on the learned reward function. Even though relatively simple approaches (e.g., rejection sampling based on reward scores) have been investigated, fine-tuning text-to-image models with the reward function remains challenging. In this work, we propose using online reinforcement learning (RL) to fine-tune text-to-image models. We focus on diffusion models, defining the fine-tuning task as an RL problem, and updating the pre-trained text-to-image diffusion models using policy gradient to maximize the feedback-trained reward. Our approach, coined DPOK, integrates policy optimization with KL regularization. We conduct an analysis of KL regularization for both RL fine-tuning and supervised fine-tuning. In our experiments, we show that DPOK is generally superior to supervised fine-tuning with respect to both image-text alignment and image quality. Our code is available at https://github.com/google-research/google-research/tree/master/dpok., Comment: NeurIPS 2023

Galactic bars can drive cold gas inflows towards the centres of galaxies. The gas transport happens primarily through the so-called bar ``dust lanes'', which connect the galactic disc at kpc scales to the nuclear rings at hundreds of pc scales much like two gigantic galactic rivers. Once in the ring, the gas can fuel star formation activity, galactic outflows, and central supermassive black holes. Measuring the mass inflow rates is therefore important to understanding the mass/energy budget and evolution of galactic nuclei. In this work, we use CO datacubes from the PHANGS-ALMA survey and a simple geometrical method to measure the bar-driven mass inflow rate onto the nuclear ring of the barred galaxy NGC~1097. The method assumes that the gas velocity in the bar lanes is parallel to the lanes in the frame co-rotating with the bar, and allows one to derive the inflow rates from sufficiently sensitive and resolved position-position-velocity diagrams if the bar pattern speed and galaxy orientations are known. We find an inflow rate of $\dot{M}=(3.0 \pm 2.1)\, \rm M_\odot\, yr^{-1}$ averaged over a time span of 40 Myr, which varies by a factor of a few over timescales of $\sim$10 Myr. Most of the inflow appears to be consumed by star formation in the ring which is currently occurring at a rate of ${\rm SFR}\simeq~1.8$-$2 \rm M_\odot\, yr^{-1}$, suggesting that the inflow is causally controlling the star formation rate in the ring as a function of time., Comment: Accepted in MNRAS

In the last decade, Convolutional Neural Network (CNN) and transformer based object detectors have achieved high performance on a large variety of datasets. Though the majority of detection literature has developed this capability on datasets such as MS COCO, these detectors have still proven effective for remote sensing applications. Challenges in this particular domain, such as small numbers of annotated objects and low object density, hinder overall performance. In this work, we present a novel augmentation method, called collage pasting, for increasing the object density without a need for segmentation masks, thereby improving the detector performance. We demonstrate that collage pasting improves precision and recall beyond related methods, such as mosaic augmentation, and enables greater control of object density. However, we find that collage pasting is vulnerable to certain out-of-distribution shifts, such as image corruptions. To address this, we introduce two simple approaches for combining collage pasting with PixMix augmentation method, and refer to our combined techniques as ColMix. Through extensive experiments, we show that employing ColMix results in detectors with superior performance on aerial imagery datasets and robust to various corruptions.

Deviation of blood flow from an optimal range is known to be associated with the initiation and progression of vascular pathologies. Important open questions remain about how the abnormal flow drives specific wall changes in pathologies such as cerebral aneurysms where the flow is highly heterogeneous and complex. This knowledge gap precludes the clinical use of readily available flow data to predict outcomes and improve treatment of these diseases. As both flow and the pathological wall changes are spatially heterogeneous, a crucial requirement for progress in this area is a methodology for co-mapping local data from vascular wall biology with local hemodynamic data. In this study, we developed an imaging pipeline to address this pressing need. A protocol that employs scanning multiphoton microscopy was designed to obtain 3D data sets for smooth muscle actin, collagen and elastin in intact vascular specimens. A cluster analysis was developed to objectively categorize the smooth muscle cells (SMC) across the vascular specimen based on SMC density. In the final step in this pipeline, the location specific categorization of SMC, along with wall thickness was co-mapped with patient specific hemodynamic results, enabling direct quantitative comparison of local flow and wall biology in 3D intact specimens., Comment: 36 pages, 5 figures

Trust is an important factor in people's interactions with AI systems. However, there is a lack of empirical studies examining how real end-users trust or distrust the AI system they interact with. Most research investigates one aspect of trust in lab settings with hypothetical end-users. In this paper, we provide a holistic and nuanced understanding of trust in AI through a qualitative case study of a real-world computer vision application. We report findings from interviews with 20 end-users of a popular, AI-based bird identification app where we inquired about their trust in the app from many angles. We find participants perceived the app as trustworthy and trusted it, but selectively accepted app outputs after engaging in verification behaviors, and decided against app adoption in certain high-stakes scenarios. We also find domain knowledge and context are important factors for trust-related assessment and decision-making. We discuss the implications of our findings and provide recommendations for future research on trust in AI., Comment: FAccT 2023

We identify giant molecular clouds (GMCs) associated with HII regions for a sample of 19 nearby galaxies using catalogs of GMCs and H regions released by the PHANGS-ALMA and PHANGS-MUSE surveys, using the overlap of the CO and H{\alpha} emission as the key criterion for physical association. We compare the distributions of GMC and HII region properties for paired and non-paired objects. We investigate correlations between GMC and HII region properties among galaxies and across different galactic environments to determine whether GMCs that are associated with HII regions have significantly distinct physical properties to the parent GMC population. We identify trends between the H{\alpha} luminosity of an HII region and the CO peak brightness and the molecular mass of GMCs that we tentatively attribute to a direct physical connection between the matched objects, and which arise independently of underlying environmental variations of GMC and HII region properties within galaxies. The study of the full sample nevertheless hides a large variability galaxy by galaxy. Our results suggests that at the ~100 pc scales accessed by the PHANGS-ALMA and PHANGS-MUSE data, pre-supernova feedback mechanisms in HII regions have a subtle but measurable impact on the properties of the surrounding molecular gas, as inferred from CO observations.