Abstract

Particulate matter (PM) is a critical air pollutant with severe health implications, yet existing stationary monitoring networks often fail to capture its complete spatial and temporal variability in urban environments. This paper presents a novel approach to city-wide air quality monitoring using low-cost sensors mounted on mobile platforms. To validate this method, a six-month field study was conducted in Hyderabad, India, deploying IoT-enabled devices on four college buses. These mobile devices capture PM concentrations, temperature, relative humidity, GPS coordinates, and vehicle speed twice daily across different parts of the city. The primary objective was to demonstrate the necessity of mobile air pollution monitoring to identify PM variability across diverse urban areas. The data revealed significant variations in PM concentrations across different parts of the city and seasons, highlighting the impact of local activities on air quality. The study examines seasonal trends, area-specific variations, and temporal patterns of PM concentrations, identifying pollution hotspots within the city. It shows how important it is to provide up-to-date, location-specific air quality information to people with pollution-related health issues.

Abstract

This report documents the program and outcomes of Dagstuhl Seminar 23302, "Software Architecture and Machine Learning". We summarize the goals and format of the seminar, results from the breakout groups, key definitions relevant to machine learning-enabled systems that were discussed, and the research roadmap that emerged from the discussions during the seminar. The report also includes the abstracts of the talks presented at the seminar and summaries of open discussions.

Abstract

Microservice architectures have gained enormous popularity due to their ability to be dynamically added/removed, replicated, and updated according to run-time needs. However, the dynamic nature of microservices introduces uncertainty, which in turn can affect the provided Quality of Service (QoS). This calls for novel service discovery mechanisms able to adapt to the variability of the QoS attributes and further perform effective service discovery and selection. To this end, this paper combines machine learning and self-adaptation techniques to perform service discovery and selection by trading off different QoS attributes. The results of our validation on a state-of-the-art microservices exemplar show that our ML-enabled approach can perform service discovery with 35% higher effectiveness with respect to existing baselines.

Abstract

This paper presents the design and development of heater control for a flexible embedded microheater (7000 um x 9000 um) for a lab-on-a-chip system. Our work focuses on advancing the PWM-based control system with the goal of reducing the system time constant. We also present the fabrication of an embedded flexible microheater in a PDMS substrate. We validate the efficacy of the flexible microheater and its control system by using it for glucose detection. A steady state temperature in the range of 40$^o$C to 120 degree C is achieved in approximately 40 s, with the heater's power consumption ranging from 0.4 W to 2 W. The novelty of the work lies in that we have made a small, flexible, low-cost chip without the use of sophisticated equipment or conventional fabrication techniques.

Abstract

Chart question answering (CQA) is a crucial area of Visual Language Understanding. However, the robustness and consistency of current Visual Language Models (VLMs) in this field remain under-explored. This paper evaluates state-of-the-art VLMs on comprehensive datasets, developed specifically for this study, encompassing diverse question categories and chart formats. We investigate two key aspects: 1) the models' ability to handle varying levels of chart and question complexity, and 2) their robustness across different visual representations of the same underlying data. Our analysis reveals significant performance variations based on question and chart types, highlighting both strengths and weaknesses of current models. Additionally, we identify areas for improvement and propose future research directions to build more robust and reliable CQA systems. This study sheds light on the limitations of current models and paves the way for future advancements in the field.

Abstract

The COVID-19 pandemic has underscored the need for low cost, scalable approaches to measuring contactless vital signs, either during initial triage at a healthcare facility or virtual telemedicine visits. Remote photoplethysmography (rPPG) can accurately estimate heart rate (HR) when applied to closeup videos of healthy volunteers in well-lit laboratory settings. However, results from such highly optimized laboratory studies may not be readily translated to healthcare settings. One significant barrier to practical application of rPPG in healthcare is accurate localization of region of interest (ROI). Clinical or telemedicine visits may involve sub-optimal lighting, movement artifacts, variable camera angle, and subject distance. This paper presents an rPPG ROI selection method based on 3D facial landmarks and patient head yaw angle. We then demonstrate robustness of this ROI selection method when coupled to Plane-Orthogonal-to-Skin (POS) rPPG method when applied to videos of patients presenting to an Emergency Department for respiratory complaints. Our primary contributions are twofold: (1) a robust ROI selection framework that adapts to real-world clinical scenarios, and (2) first unrestricted rPPG dataset collected from emergency ward settings, addressing critical gaps between controlled laboratory conditions and real-world clinical environments. Our results demonstrate effectiveness of our proposed approach in improving accuracy and robustness of rPPG in a challenging clinical environment.

Abstract

Supervised machine learning methods for geological mapping via remote sensing face limitations due to the scarcity of accurately labelled training data that can be addressed by unsupervised learning, such as dimensionality reduction and clustering. Dimensionality reduction methods have the potential to play a crucial role in improving the accuracy of geological maps. Although conventional dimensionality reduction methods may struggle with nonlinear data, unsupervised deep learning models such as autoencoders can model non-linear relationships. Stacked autoencoders feature multiple interconnected layers to capture hierarchical data representations useful for remote sensing data. We present an unsupervised machine learning-based framework for processing remote sensing data using stacked autoencoders for dimensionality reduction and k-means clustering for mapping geological units. We use Landsat 8, ASTER, and Sentinel-2 datasets to evaluate the framework for geological mapping of the Mutawintji region in Western New South Wales, Australia. We also compare stacked autoencoders with principal component analysis (PCA) and canonical autoencoders. Our results reveal that the framework produces accurate and interpretable geological maps, efficiently discriminating rock units. The results reveal that the combination of stacked autoencoders with Sentinel-2 data yields the best performance accuracy when compared to other combinations. We find that stacked autoencoders enable better extraction of complex and hierarchical representation of the input data when compared to canonical autoencoders and PCA. We also find that the generated maps align with prior geological knowledge of the study area while providing novel insights into geological structures.

Abstract

Crack segmentation plays a crucial role in ensuring the structural integrity and seismic safety of civil structures. However, existing crack segmentation algorithms encounter challenges in maintaining accuracy with domain shifts across datasets. To address this issue, we propose a novel deep network that employs incremental training with unsupervised domain adaptation (UDA) using adversarial learning, without a significant drop in accuracy in the source domain. Our approach leverages an encoder-decoder architecture, consisting of both domain-invariant and domain-specific parameters. The encoder learns shared crack features across all domains, ensuring robustness to domain variations. Simultaneously, the decoder's domain-specific parameters capture domain-specific features unique to each domain. By combining these components, our model achieves improved crack segmentation performance. Furthermore, we introduce BuildCrack, a new crack dataset comparable to sub-datasets of the well-established CrackSeg9K dataset in terms of image count and crack percentage. We evaluate our proposed approach against state-of-the-art UDA methods using different sub-datasets of CrackSeg9K and our custom dataset. Our experimental results demonstrate a significant improvement in crack segmentation accuracy and generalization across target domains compared to other UDA methods - specifically, an improvement of 0.65 and 2.7 mIoU on source and target domains respectively. Code, models, and dataset will be made available.

Abstract

Polynomial dynamical systems (i.e. dynamical systems with polynomial right hand side) are ubiquitous in applications, especially as models of reaction networks and interaction networks. The properties of general polynomial dynamical systems can be very difficult to analyze, due to nonlinearity, bifurcations, and the possibility for chaotic dynamics. On the other hand, toric dynamical systems are polynomial dynamical systems that appear naturally as models of reaction networks, and have very robust and stable properties. A disguised toric dynamical system is a polynomial dynamical system generated by a reaction network N and some choice of positive parameters, such that (even though it may not be toric with respect to N) it has a toric realization with respect to some network N′. Disguised toric dynamical systems enjoy all the robust stability properties of toric dynamical systems. In this paper, we study a larger set of dynamical systems where the rate constants are allowed to take both positive and negative values. More precisely, we analyze the R-disguised toric locus of a reaction network N, i.e., the subset in the space rate constants (positive or negative) of N for which the corresponding polynomial dynamical system is disguised toric. We focus especially on finding a lower bound on the dimension of the R-disguised toric locus.

Abstract

Visual servoing for the development of autonomous robotic systems capable of administering UltraSound (US) guided regional anesthesia requires real-time segmentation of nerves, needle tip localization and needle trajectory extrapolation. First, we recruited 227 patients to build a large dataset of 41,000 anesthesiologist annotated images from US videos of brachial plexus nerves and developed models to localize nerves in the US images. Generalizability of the best suited model was tested on the datasets constructed from separate US scanners. Using these nerve segmentation predictions, we define automated anesthesia needle targets by fitting an ellipse to the nerve contours. Next, we developed an image analysis tool to guide the needle toward their targets. For the segmentation of the needle, a natural RGB pre-trained neural network was first fine-tuned on a large US dataset for domain transfer and then adapted for the needle using a small dataset. The segmented needle’s trajectory angle is calculated using Radon transformation and the trajectory is extrapolated from the needle tip. The intersection of the extrapolated trajectory with the needle target guides the needle navigation for drug delivery. The needle trajectory’s average error was within acceptable range of 5 mm as per experienced anesthesiologists. The entire dataset has been released publicly for further study by the research community at https://github.com/Regional-US/