Om Chabra’s research while affiliated with University of Illinois Urbana-Champaign and other places

Constructing real-world data-to-insight pipelines often involves data extraction from data lakes, data integration across heterogeneous data sources, and diverse operations from data cleaning to analysis. The design and implementation of data science pipelines require domain knowledge, technical expertise, and even project-specific insights. AI systems have shown remarkable reasoning, coding, and understanding capabilities. However, it remains unclear to what extent these capabilities translate into successful design and execution of such complex pipelines. We introduce KRAMABENCH: a benchmark composed of 104 manually-curated real-world data science pipelines spanning 1700 data files from 24 data sources in 6 different domains. We show that these pipelines test the end-to-end capabilities of AI systems on data processing, requiring data discovery, wrangling and cleaning, efficient processing, statistical reasoning, and orchestrating data processing steps given a high-level task. Our evaluation tests 5 general models and 3 code generation models using our reference framework, DS-GURU, which instructs the AI model to decompose a question into a sequence of subtasks, reason through each step, and synthesize Python code that implements the proposed design. Our results on KRAMABENCH show that, although the models are sufficiently capable of solving well-specified data science code generation tasks, when extensive data processing and domain knowledge are required to construct real-world data science pipelines, existing out-of-box models fall short. Progress on KramaBench represents crucial steps towards developing autonomous data science agents for real-world applications. Our code, reference framework, and data are available at https://github.com/mitdbg/KramaBench.

In this paper, we present a new city-scale decentralized mesh network system suited for disaster recovery and emergencies. When wide-area connectivity is unavailable or significantly degraded, our system, MapMesh, enables static access points and mobile devices equipped with Wi-Fi in a city to route packets via each other for intra-city connectivity and to/from any nodes that might have Internet access, e.g., via satellite. The chief contribution of our work is a new routing protocol that scales to millions of nodes, a significant improvement over prior work on wireless mesh and mobile ad hoc networks. Our approach uses detailed information about buildings from widely available maps--data that was unavailable at scale over a decade ago, but is widely available now--to compute paths in a scalable way.

The Stingray sensor system is a 15-camera optical array dedicated to the nightly astrometric and photometric survey of the geosynchronous Earth orbit (GEO) belt visible above Tucson, Arizona. The primary scientific goal is to characterize GEO and near-GEO satellites based on their observable properties. This system is completely autonomous in both data acquisition and processing, with human oversight reserved for data quality assurance and system maintenance. The 15 ZWO ASI1600MM Pro cameras are mated to Sigma 135 mm f/1.8 lenses and are controlled simultaneously by four separate computers. Each camera is fixed in position and observes a 7.6-by-5.8-degree portion of the GEO belt, for a total of a 114-by-5.8-degree field of regard. The GAIA DR2 star catalog is used for image astrometric plate solution and photometric calibration to GAIA G magnitudes. There are approximately 200 near-GEO satellites on any given night that fall within the Stingray field of regard, and all those with a GAIA G magnitude brighter than approximately 15.5 are measured by the automated data reduction pipeline. Results from an initial one-month survey show an aggregate photometric uncertainty of 0.062 ± 0.008 magnitudes and astrometric accuracy consistent with theoretical sub-pixel centroid limits. Provided in this work is a discussion of the design and function of the system, along with verification of the initial survey results.

Low Earth Orbit (LEO) satellite constellations are enhancing IoT device connectivity with cost-effective, easily deployable picosatellites. These IoT satellite constellations bypass the need for Earth-based gateways, offering global coverage. This article delves into communication challenges encountered when deploying an IoT satellite network. Shifting the classic perspective on Doppler from a hindrance to an advantage, we present novel techniques leveraging Doppler shift for reliable packet detection and decoding, particularly in noisy environments. Integrated into our system, Spectrumize [1], these techniques offer a substantial performance boost compared to conventional methods.

(16) Psyche is the largest M-type asteroid in the main belt and the target of the NASA Discovery-class Psyche mission. Despite gaining considerable interest in the scientific community, Psyche's composition and formation remain unconstrained. Originally, Psyche was considered to be almost entirely composed of metal due to its high radar albedo and spectral similarities to iron meteorites. More recent telescopic observations suggest the additional presence of low-Fe pyroxene and exogenic carbonaceous chondrites on the asteroid's surface. To better understand the abundances of these additional materials, we investigated visible near-infrared (0.35–2.5 μ m) spectral properties of three-component laboratory mixtures of metal, low-Fe pyroxene, and carbonaceous chondrite. We compared the band depths and spectral slopes of these mixtures to the telescopic spectrum of (16) Psyche to constrain material abundances. We find that the best matching mixture to Psyche consists of 82.5% metal, 7% low-Fe pyroxene, and 10.5% carbonaceous chondrite by weight, suggesting that the asteroid is less metallic than originally estimated (∼94%). The relatively high abundance of carbonaceous chondrite material estimated from our laboratory experiments implies the delivery of this exogenic material through low velocity collisions to Psyche's surface. Assuming that Psyche's surface is representative of its bulk material content, our results suggest a porosity of 35% to match recent density estimates.

(16) Psyche is the largest M-type asteroid in the main belt and the target of the NASA Discovery-class Psyche mission. Despite gaining considerable interest in the scientific community, Psyche's composition and formation remain unconstrained. Originally, Psyche was considered to be almost entirely composed of metal due to its high radar albedo and spectral similarities to iron meteorites. More recent telescopic observations suggest the additional presence of low-Fe pyroxene and exogenic carbonaceous chondrites on the asteroid's surface. To better understand the abundances of these additional materials, we investigated visible near-infrared (0.35 - 2.5 micron) spectral properties of three-component laboratory mixtures of metal, low-Fe pyroxene, and carbonaceous chondrite. We compared the band depths and spectral slopes of these mixtures to the telescopic spectrum of (16) Psyche to constrain material abundances. We find that the best matching mixture to Psyche consists of 82.5% metal, 7% low-Fe pyroxene, and 10.5% carbonaceous by weight, suggesting that the asteroid is less metallic than originally estimated (~94%). The relatively high abundance of carbonaceous chondrite material estimated from our laboratory experiments implies the delivery of this exogenic material through low velocity collisions to Psyche's surface. Assuming that Psyche's surface is representative of its bulk material content, our results suggest a porosity of 35% to match recent density estimates.