UMBC Computer Science and Electrical Engineering Department

Permanent URI for this collectionhttp://hdl.handle.net/11603/50

The Computer Science and Electrical Engineering Department aims to maintain a program of excellence in teaching, research, and service for all of its programs. At the undergraduate level, we will provide students with a firm foundation of both the theory and practice of Computer Science and Computer Engineering. Our curricula also give students the social, ethical, and liberal education needed to make significant contributions to society. Students receiving a bachelor’s degree are ready to enter the work force as productive computer scientists or computer engineers, or to continue their education at the graduate or professional level.

At the graduate level, we are committed to developing the research and professional capabilities of students in Computer Science, Computer Engineering, Electrical Engineering and Cybersecurity. Our programs provide a deeper mastery of the basics of these fields, as well as opportunities to collaborate on leading-edge research with our faculty. Our faculty are engaged in both practical and theoretical research, often in partnership with government agencies, private industry and non-governmental organizations. The aim of this research is to advance knowledge within our disciplines and also to contribute to solving problems faced by our society.

Browse

Now showing 1 - 20 of 2213

PhysiFi: WiFi Sensing for Monitoring Therapeutic Robotic Systems
(2025) Akpabio, Wonder; Bulut, Eyuphan
Patients recovering from limb-impairing strokes require consistent and precise physical therapy (PT) to regain mobility and functionality. Autonomous rehabilitation robots are increasingly adopted during recovery, offering a scalable solution to reduce the burden on physical therapists while assisting patients in performing prescribed exercises accurately. However, the effectiveness of these treatments often relies on professional supervision to ensure patients follow the robot’s movements properly, which could be challenging considering the ongoing shortage of physical therapists. Current PT monitoring systems primarily rely on camera-based technologies, which usually raise concerns due to potential privacy violations and high deployment costs, or wearable devices that are intrusive and uncomfortable for patients. To address these limitations, we propose PhysiFi, a novel approach that leverages ubiquitous WiFi signals available in most indoor environments, such as homes, rehabilitation centers, and assisted living facilities. By analyzing Channel State Information (CSI) from ambient WiFi signals and employing deep learning models, PhysiFi can track and recognize exercises performed by patients with rehabilitation robots. Our experiments demonstrate that PhysiFi can accurately identify prescribed exercises and evaluate whether patients are following the robot’s movements correctly, providing a non-intrusive, privacy-preserving, and costeffective alternative for monitoring physical therapy sessions.
Rainfall Frequency Analysis Based on Long-Term High-Resolution Radar Rainfall Fields: Spatial Heterogeneities and Temporal Nonstationarities
(AGU, 2024) Smith, James A.; Baeck, Mary Lynn; Miller, Andrew; Claggett, Elijah L.
Rainfall frequency analysis methods are developed and implemented based on high-resolution radar rainfall data sets, with the Baltimore metropolitan area serving as the principal study region. Analyses focus on spatial heterogeneities and time trends in sub-daily rainfall extremes. The 22-year radar rainfall data set for the Baltimore study region combines reflectivity-based rainfall fields during the period from 2000 to 2011 and polarimetric rainfall fields for the period from 2012 to 2021. Rainfall frequency analyses are based on non-stationary formulations of peaks-over-threshold and annual peak methods. Increasing trends in short-duration rainfall extremes are inferred from both peaks-over-threshold and annual peak analyses for the period from 2000 to 2021. There are pronounced spatial gradients in short-duration rainfall extremes over the study region, with peak values of rainfall between Baltimore City and Chesapeake Bay. Spatial gradients in 100-year, 1 hr rainfall over 20 km length scale are comparable to time trends over 20 years. Rainfall analyses address the broad challenge of assessing changing properties of short-duration rainfall in urban regions. Analyses of high-resolution rainfall fields show that sub-daily rainfall extremes are only weakly related to daily extremes, pointing to difficulties in inferring climatological properties of sub-daily rainfall from daily rainfall analyses. Changing measurement properties are a key challenge for application of radar rainfall data sets to detection of time trends. Mean field bias correction of radar rainfall fields using rain gauge observations is an important tool for improving radar rainfall fields and provides a useful tool for addressing problems associated with changing radar measurement properties.
Towards robust visual understanding: A paradigm shift in computer vision from recognition to reasoning
(Wiley, 2024) Gokhale, Tejas
Models that learn from data are widely and rapidly being deployed today for real-world use, but they suffer from unforeseen failures that limit their reliability. These failures often have several causes such as distribution shift; adversarial attacks; calibration errors; scarcity of data and/or ground-truth labels; noisy, corrupted, or partial data; and limitations of evaluation metrics. But many failures also occur because many modern AI tasks require reasoning beyond pattern matching and such reasoning abilities are difficult to formulate as data-based input–output function fitting. The reliability problem has become increasingly important under the new paradigm of semantic “multimodal” learning. In this article, I will discuss findings from our work to provide avenues for the development of robust and reliable computer vision systems, particularly by leveraging the interactions between vision and language. This article expands upon the invited talk at AAAI 2024 and covers three thematic areas: robustness of visual recognition systems, open-domain reliability for visual reasoning, and challenges and opportunities associated with generative models in vision.
Realizing 3D Visualization U-sing Crossed-Beam Volumetric Displays
(ACM, 1998-08-01) Ebert, David; Bedwell, Edward; Maher, Stephen; Smoliar, Laura; Downing, Elizabeth
P-1694. Estimating the Use of Newer Beta-lactam Gram-negative Targeted Antibiotics Across Greater than 700 Hospitals in the United States
(Oxford University Press, 2025-01-29) Harris, Anthony; Goodman, Katherine E.; Pineles, Lisa; Walker, Morgan; Bork, Jacqueline T.; Heil, Emily L.; Claeys, Kimberly C.; Brooks, Justin; Kadri, Sameer S.; Maron, Brad; Baghdadi, Jonathan
Newer broad-spectrum beta-lactam antibiotics (cefiderocol, ceftazidime-avibactam, ceftolozane-tazobactam, imipenem-relebactam, meropenem-vaborbactam) have been introduced. Ceftazidime-avibactam and ceftolozane-tazobactam predominated newer antibiotic use through 2021; however, less is known about current use patterns and for what clinical indications, in inpatient settings. The aim of this study was to describe the use of these newer antibiotics across a large national cohort. We performed a retrospective cohort study of adults discharged from June 2022 through May 2023 among hospitals in the Premier Healthcare Database. Antibiotic utilization was ascertained from daily charge data, and clinical indication(s) were inferred from ICD-10 diagnosis codes. A stratified analysis was performed among patients receiving >3 days of antibiotic therapy to understand when these antibiotics are being used as definitive therapy. Across 832 hospitals, 3,890,557 admissions (61.9% of all admissions) included receipt of antibiotics. In 8,655 admissions (0.2% of antibiotic-prescribing admissions), newer antibiotics were prescribed (see Table 1 for patient demographic distribution). Ceftolozane-tazobactam was prescribed in 4157 (48%), 3660 (42.3%) ceftazidime-avibactam, 1060 (12.2%) cefiderocol, 456 (5.3%) meropenem-vaborbactam, and 99 (1.1%) imipenem-relebactam. Most patients who received newer antibiotics (69%, n=5,998) received them for greater than 3 days. Of these, ceftolozane-tazobactam was prescribed in 2607 admissions (43.5%), 2505 (41.7%) ceftazidime-avibactam, 832 (13.9%) cefiderocol, 332 (5.5%)meropenem-vaborbactam, and 68 (1.1%) imipenem-relebactam for greater than 3 days. Sepsis was the most common clinical indication for receipt of newer agents, followed by urinary tract infection (Table 2). Ceftazidime-avibactam and ceftolozane-tazobactam remain the most frequently prescribed new antibiotics, with uptake of subsequently approved agents trailing. Most of the prescribing of these antibiotics are for sepsis.
Neuro-Symbolic AI for Deep Analysis of Social Media Big Data
(2024-09-13) Khandelwal, Vedant; Gaur, Manas; Kursuncu, Ugur; Shalin, Valerie; Sheth, Amit P.
This tutorial introduces a neuro-symbolic AI framework to analyze big data from social media platforms. Integrating human-curated knowledge through symbolic AI with the pattern recognition capabilities of neural networks enhances the adaptability and efficiency of traditional neural network approaches. Knowledge-guided zero-shot learning techniques enable swift adaption to new linguistic contexts and emerging events [6]. Participants will explore how to design, develop, and utilize these models in specific domains, such as public health surveillance that require dynamic adaptation to new terminologies. This session The tutorial aims to equip attendees with practical skills and a deep understanding of how to apply neuro-symbolic AI to manage and analyze large-scale social media datasets effectively.
Open-Source Large Language Models in Radiology: A Review and Tutorial for Practical Research and Clinical Deployment
(RSNA, 2025-01-28) Savage, Cody H.; Kanhere, Adway; Parekh, Vishwa; Langlotz, Curtis P.; Joshi, Anupam; Huang, Heng; Doo, Florence X.
Integrating large language models (LLMs) into health care holds substantial potential to enhance clinical workflows and care delivery. However, LLMs also pose serious risks if integration is not thoughtfully executed, with complex challenges spanning accuracy, accessibility, privacy, and regulation. Proprietary commercial LLMs (eg, GPT-4 [OpenAI], Claude 3 Sonnet and Claude 3 Opus [Anthropic], Gemini [Google]) have received much attention from researchers in the medical domain, including radiology. Interestingly, open-source LLMs (eg, Llama 3 and LLaVA-Med) have received comparatively little attention. Yet, open-source LLMs hold several key advantages over proprietary LLMs for medical institutions, hospitals, and individual researchers. The wider adoption of open-source LLMs has been slower, perhaps in part due to the lack of familiarity, accessible computational infrastructure, and community-built tools to streamline their local implementation and customize them for specific use cases. Thus, this article provides a tutorial for the implementation of open-source LLMs in radiology, including examples of commonly used tools for text generation and techniques for troubleshooting issues with prompt engineering, retrieval-augmented generation, and fine-tuning. Implementation-ready code for each tool is provided at https://github.com/UM2ii/Open-Source-LLM-Tools-for-Radiology. In addition, this article compares the benefits and drawbacks of open-source and proprietary LLMs, discusses the differentiating characteristics of popular open-source LLMs, and highlights recent advancements that may affect their adoption.
Gaps in U.S. livestock data are a barrier to effective environmental and disease management
(AAS, 2025-02-11) Logsdon Muenich, Rebecca; Aryal, Sanskriti; Ashworth, Amanda J; Bell, Michelle L; Boudreau, Melanie R; Cunningham, Stephanie A; Flynn, K Colton; Hamilton, Kerry A; Liu, Ting; Mashtare, Michael L; Nelson, Natalie G; Rashid, Barira; Saha, Arghajeet; Schaffer-Smith, Danica; Showalter, Callie; Tchamdja, Aureliane; Thompson, Jada
Livestock are a critical part of our food systems, yet their abundance globally has been cited as a driver of many environmental and human health concerns. Issues such as soil, water, and air pollution, greenhouse gas emissions, aquifer depletion, antimicrobial resistance genes, and zoonotic disease outbreaks have all been linked to livestock operations. While many studies have examined these issues at depth at local scales, it has been difficult to complete studies at regional or national scales due to the dearth of livestock data, hindering pollution mitigation or response time for tracing and monitoring disease outbreaks. In the U.S. the National Agricultural Statistics Service completes a Census once every 5 years that includes livestock, but data are only available at the county level leaving little inference that can be made at such a coarse spatiotemporal scale. While other data exist through some regulated permitting programs, there are significant data gaps in where livestock are raised, how many livestock are on site at a given time, and how these livestock and, importantly, their waste emissions, are managed. In this perspective, we highlight the need for better livestock data, then discuss the accessibility and key limitations of currently available data. We then feature some recent work to improve livestock data availability through remote-sensing and machine learning, ending with our takeaways to address these data needs for the future of environmental and public health management.
Probing an auxiliary laser to tune the repetition rate of a soliton microcomb
(Optica, 2025-02-15) Mahmood, Tanvir; Cahill, James P.; Sykes, Patrick; Courtright, Logan; Wu, Lue; Vahala, Kerry J.; Menyuk, Curtis; Zhou, Weimin
We demonstrate that it is possible to linearly tune the repetition rate of a bright soliton comb that is generated using an Si3N4 microring resonator by linearly varying the frequency of an auxiliary heater laser. Hence, the auxiliary laser can be utilized as a linear active feedback element for stabilizing the repetition rate. We investigated the potential of the auxiliary laser as an actuator of the soliton repetition rate by varying the auxiliary laser frequency at different modulation rates. Within the modulation bandwidth of the laser, we find that the variation ratio, defined as the ratio of the change in the repetition rate to the change in the laser frequency, remains unchanged. This variation ratio also quantifies the correlation between the frequency drift of the auxiliary laser and the repetition rate phase noise and makes it possible to examine the impact of frequency drift on the attainable phase noise performance of the soliton microcomb. For our setup, we find that the repetition rate phase noise of the microcomb below a 1-kHz offset from the carrier is dominated by the frequency drift of the auxiliary laser, which emphasizes the importance of deploying an inherently low-phase-noise laser when auxiliary laser heating technique is utilized.
Accurate and Interpretable Radar Quantitative Precipitation Estimation with Symbolic Regression
(IEEE, 2025-01-16) Zhang, Olivia; Grissom, Brianna; Pulido, Julian; Munoz-Ordaz, Kenia; He, Jonathan; Cham, Mostafa; Jing, Haotong; Qian, Weikang; Wen, Yixin; Wang, Jianwu
Accurate quantitative precipitation estimation (QPE) is essential for managing water resources, monitoring flash floods, creating hydrological models, and more. Traditional methods of obtaining precipitation data from rain gauges and radars have limitations such as sparse coverage and inaccurate estimates for different precipitation types and intensities. Symbolic regression, a machine learning method that generates mathematical equations fitting the data, presents a unique approach to estimating precipitation that is both accurate and interpretable. Using WSR-88D dual-polarimetric radar data from Oklahoma and Florida over three dates, we tested symbolic regression models involving genetic programming and deep learning, symbolic regression on separate clusters of the data, and the incorporation of knowledge-based loss terms into the loss function. We found that symbolic regression is both accurate in estimating rainfall and interpretable through learned equations. Accuracy and simplicity of the learned equations can be slightly improved by clustering the data based on select radar variables and by adjusting the loss function with knowledge-based loss terms. This research provides insights into improving QPE accuracy through interpretable symbolic regression methods
Applause: A Learning Tool for Low-Resource Languages
(2014) Wolfe, Nikolas; Vemuri, Vinay Vyas; Martin, Lara J.; Metze, Florian; Black, Alan W.
In this paper, we describe the concept of an interactive tool which can be employed to build a dialog system to facilitate pronunciation training in situations where only a few minutes of speech in the target language are available. We leverage recent advances in low-resource speech processing, and envision a tool which will help organizations working in the developing world to quickly create initial training lessons for new languages. Development will be possible using mobile devices alone, and without the requirement of significant technical skills or language-specific information. If users are ultimately able to acquire at least rudimentary proficiency in a low-resource language or dialect using our automatic system, they should find it much easier to communicate, establish trust, and build rapport with the local population than without such support. In this paper, we present the operational principles of our approach, describe a proof-of-concept implementation that we are currently developing, and summarize ideas for evaluation at the technical, language-learning, and user interface levels.
Informedia@TRECVID 2014 MED and MER
(2014) Yu, Shoou-I.; Jiang, Lu; Xu, Zhongwen; Lan, Zhenzhong; Xu, Shicheng; Chang, Xiaojun; Martin, Lara J.
We report on our system used in the TRECVID 2014 Multimedia Event Detection (MED) and Multimedia Event Recounting (MER) tasks. On the MED task, the CMU team achieved leading performance in the Semantic Query (SQ), 000Ex, 010Ex and 100Ex settings. Furthermore, SQ and 000Ex runs are significantly better than the submissions from the other teams. We attribute the good performance to 4 main components: 1) large-scale semantic concept detectors trained on video shots for SQ/000Ex systems, 2) better features such as improved trajectories and deep learning features for 010Ex/100Ex systems, 3) a novel Multistage Hybrid Late Fusion method for 010Ex/100Ex systems and 4) improved reranking methods for Pseudo Relevance Feedback for 000Ex/010Ex systems. On the MER task, our system utilizes a subset of features and detection results from the MED system from which the recounting is then generated. Recounting evidence is presented by selecting the most likely concepts detected in the salient shots of a video. Salient shots are detected by searching for shots which have high response when predicted by the video level event detector.
Identifying Student Leaders from MOOC Discussion Forums through Language Influence
(ACL Anthology, 2014-10) Moon, Seungwhan; Potdar, Saloni; Martin, Lara J.
Identifying and understanding the motivations of student leaders from Massively Open Online Course (MOOC) discussion forums provides the key to making the online learning environment engaging, collaborative, and instructive. In this paper, we propose to identify student leaders solely based on textual features, or specifically by analyzing how they influence other students’ language. We propose an improved method of measuring language accommodation based on people’s choice of words given a semantic topic of interest, and show that student leaders indeed coordinate other students’ language usage. We also show that our proposed method can successfully distinguish student leaders from the two MOOC discussion forum datasets.
A multisensory non-invasive system for laughter analysis
(2014) Cosentino, Sarah; Burger, Susanne; Martin, Lara J.; Metze, Florian; Kishi, Tatsuhiro; Hashimoto, Kenji; Sessa, Salvatore; Zecca, Massimiliano; Takanishi, Atsuo
Laughter is an important non-verbal human social signal. Clarifying the mechanism of laughing would be useful in a variety of studies on health or sociology. In this paper we introduce a non-invasive multisensory system for real-time laughter detection and analysis. We focus only on the audio laughter recognition, present the preliminary results we obtained, and discuss the possible application of this system in the medical field.
A methodology for using crowdsourced data to measure uncertainty in natural speech
(IEEE, 2014-12) Martin, Lara J.; Stone, Matthew; Metze, Florian; Mostow, Jack
People sometimes express uncertainty unconsciously in order to add layers of meaning on top of their speech, conveying doubts about the accuracy of the information they are trying to communicate. In this paper, we propose a methodology for annotating uncertainty, which is usually a subjective and expensive process, by using crowdsourcing. In our experiment, we used an online database which consists of colors that more than 200,000 users have named. Based on the amount of unique names that users have given each color, an entropy value was calculated to represent the uncertainty level of the color. A model, which performed better than chance, was created to predict whether or not the color that the participant was describing was ambiguous or borderline, given certain prosodic cues of their speech when asked to name the color verbally. Using crowdsourced data can greatly streamline the process of annotating uncertainty, but our methods have yet to be tested in other domains besides color. By using methods such as ours to measure prosodic attributes of uncertainty, it should be possible to increase the accuracy of voice search.
CMU Informedia@TRECVID 2015 MED
(NIST, 2015) Yu, Shoou-I.; Jiang, Lu; Xu, Zhongwen; Lan, Zhenzhong; Xu, Shicheng; Chang, Xiaojun; Li, Xuanchong; Mao, Zexi; Gan, Chuang; Miao, Yajie; Du, Xingzhong; Cai, Yang; Martin, Lara J.; Wolfe, Nikolas; Li, Huan; Lin, Ming; Ma, Zhigang; Yang, Yi; Meng, Deyu; Shan, Shiguang; Duygulu, Pinar; Burger, Susanne; Metze, Florian; Singh, Rita; Raj, Bhiksha; Mitamura, Teruko; Stern, Richard; Hauptmann, Alexander
We report on our system used in the TRECVID 2015 Multimedia Event Detection (MED) task. On the MED task, the CMU team submitted runs in the Semantic Query (SQ) and 10Ex settings. The proposed system is essentially the same as our MED 2014 system.
Improvisational Storytelling Agents
Martin, Lara J.; Ammanabrolu, Prithviraj; Wang, Xinyu; Singh, Shruti; Harrison, Brent; Dhuliawala, Murtaza; Tambwekar, Pradyumna; Mehta, Animesh; Arora, Richa; Dass, Nathan; Purdy, Chris; Riedl, Mark O.
The problem of improvisational story generation involves one or more agents collaborating in order to create a story without any advance notice of topic. We present a pipeline for an artificial agent that is capable of improvisational storytelling while collaborating with a human agent. Starting with story corpora, we “eventify” sentences, which creates a simplified and abstracted representation. The rest of the pipeline–the agent’s response–is broken into three parts: generating successive events (event-to-event), translating of events back into natural language (event-to-sentence), and plugging the specifics of the story back into the generated sentences (slot filling). We discuss techniques for each of these sub-problems.
Utterance classification in speech-to-speech translation for zero-resource languages in the hospital administration domain
(IEEE, 2015-12) Martin, Lara J.; Wilkinson, Andrew; Miryala, Sai Sumanth; Robison, Vivian; Black, Alan W.
Although substantial progress has been achieved in speech-to-speech translation systems over the last few years, such systems still require that the speech be written in some appropriate orthography. As speech may differ greatly from the standardized written form of a language, it can be non-trivial to collect written data when there is no standard way for it to be represented. This project addresses the problem from the other end and expects that speech alone is available in the target language, and that no (standard or non-standard) orthography exists. It, therefore, treats the acoustic representation of the language as primary and uses language-independent methods to produce a phonetically-related symbolic representation that is then used in the translation system. Thus, the speech translation system is created for the target language as defined by the recording of that language rather than some body of orthographic transcripts. In this work, we are creating an application called APT (Acoustic Patient Translator), which uses a novel scheme of speech recognition and translation within a targeted domain. By working with a set of predefined sentences appropriately chosen to fit a scenario, we use utterance classification as a speech recognition algorithm. The utterance classification is achieved using cross-lingual, language-independent phonetic labeling. Since we are working with a set of select phrases, the translation part is trivial. We are concentrating on communication with hospital staff, such as scheduling a doctor's appointment, as our domain. In addition to English, we also run experiments on Tamil.
Event representations for automated story generation with deep neural nets
(ACM, 2018-02-02) Martin, Lara J.; Ammanabrolu, Prithviraj; Wang, Xinyu; Hancock, William; Singh, Shruti; Harrison, Brent; Riedl, Mark O.
Automated story generation is the problem of automatically selecting a sequence of events, actions, or words that can be told as a story. We seek to develop a system that can generate stories by learning everything it needs to know from textual story corpora. To date, recurrent neural networks that learn language models at character, word, or sentence levels have had little success generating coherent stories. We explore the question of event representations that provide a mid-level of abstraction between words and sentences in order to retain the semantic information of the original data while minimizing event sparsity. We present a technique for preprocessing textual story data into event sequences. We then present a technique for automated story generation whereby we decompose the problem into the generation of successive events (event2event) and the generation of natural language sentences from events (event2sentence). We give empirical results comparing different event representations and their effects on event successor generation and the translation of events to natural language.
Dungeons and DQNs: Toward Reinforcement Learning Agents that Play Tabletop Roleplaying Games
(2018) Martin, Lara J.; Sood, Srijan; Riedl, Mark O.
Game playing has been an important testbed for artificial intelligence. Board games, first-person shooters, and real-time strategy games have well-defined win conditions and rely on strong feedback from a simulated environment. Text adventures require natural language understanding to progress through the game but still have an underlying simulated environment. In this paper, we propose tabletop roleplaying games as a challenge due to an infinite action space, multiple (collaborative) players and models of the world, and no explicit reward signal. We present an approach for reinforcement learning agents that can play tabletop roleplaying games.

Browse

Recent Submissions