Center for Information Technology

Proxemics theory argues for the importance of people regulating the distance they maintain from each other, allowing for the construction of a personal space boundary which, when violated, generates discomfort. This distance is regulated by a complex analysis mechanism that involves the fusion of cultural elements with a set of non-verbal factors, such as kinesics, haptics, and visual cues. Human comfort distance should also be considered in interactions with robotic agents. In the workplace, humans tend to tolerate variations in classic social distances (intimate, personal, social, and public) when engaging in common activities with clearly defined work-related goals. This study investigates the hypothesis that comfort distance with a robotic agent is also flexible when a joint action is underway. To test this hypothesis, an experiment involving interaction between a human participant and a collaborative robot (cobot) was conducted, in which a sense of shared interest was established through a card game. In addition to quantitative data, such as robot distance, the experiment also generates qualitative data through participants’ responses to a questionnaire about their impressions of the interaction. The findings are particularly relevant to industrial settings, where the use of cobot is increasingly in demand.

Persistent knowledge is essential for propagating the learning health system (LHS) cycle. Integral to the cycle are iterative transformations of data into knowledge. However, human efforts to undertake these transformations are increasingly challenged when dealing with larger data scales and complexities. Data sets within repositories and archives are often underutilized unless specifically requested for research programs. Specialized software algorithms (agents) can use existing knowledge for learning tasks, explore their environment, discover and create goals, and interact with humans. This paper examines the potential role of software intelligence for autonomous transformations of data and knowledge. Agents can perform various goal-directed tasks. Multi-agent systems can be utilized for data collection, description, preparation, modeling, and knowledge-mining tasks. Knowledge representation, ontologies, semantic web standards, knowledge bases, and graphs can lead to a higher level of directed learning. Agents can develop reasoning abilities and self-generate goals by leveraging semantic relationships between various datasets. A conceptual framework for an intelligent biomedical platform (IBP) is proposed. The IBP comprises four layers: infrastructure (IS), user interface (UI), coordination system (CS), and data and knowledge (DK). It also integrates a network of multi-agent systems for clinical decision-making and knowledge-mining tasks. Intelligence in the platform results from the interaction of the IS, UI, CS, and DK agents. These agents can implement multiple inferential steps using the data and knowledge within accessible repositories. Large language models can be integrated with various knowledge resources and domain-specific databases, thereby improving the accuracy of results. An IBP supported by a multi-agent system can enhance the autonomous transformation of data and knowledge. Including software intelligence within current repositories and archives enhances data reuse and the generation of new knowledge. With the addition of software reasoning capabilities in biomedical platforms, the LHS cycle can be efficiently propagated to aid in newer biomedical discoveries.

With the rapid advancements in multi-modal large language models (MLLMs), connectors play a pivotal role in bridging diverse modalities and enhancing model performance. However, the design and evolution of connectors have not been comprehensively analyzed, leaving gaps in understanding how these components function and hindering the development of more powerful connectors. In this survey, we systematically review the current progress of connectors in MLLMs and present a structured taxonomy that categorizes connectors into atomic operations (mapping, compression, mixture of experts) and holistic designs (multi-layer, multi-encoder, multi-modal scenarios), highlighting their technical contributions and advancements. Furthermore, we discuss several promising research frontiers and challenges, including high-resolution input, dynamic compression, guide information selection, combination strategy, and interpretability. This survey is intended to serve as a foundational reference and a clear roadmap for researchers, providing valuable insights into the design and optimization of next-generation connectors to enhance the performance and adaptability of MLLMs.

Current studies regarding the secondary use of electronic health records (EHR) predominantly rely on domain expertise and existing medical knowledge. A powerful representation approach can unleash the potential of discovering new medical patterns underlying the EHR. Here, we introduce an unsupervised method for embedding high-dimensional EHR data at the patient level to characterize heterogeneity in complex diseases and identify novel disease patterns linked to disparities in clinical outcomes. We applied this approach to 34,851 unique medical codes across 1,046,649 longitudinal patient events, including 102,740 patients in the Electronic Medical Records and GEnomics (eMERGE) Network. The model achieved strong predictive performance in predicting future disease (median AUROC = 0.87 within one year) and bulk phenotyping (median AUROC = 0.84). Notably, these patient embeddings revealed diverse comorbidity profiles and health outcomes, including distinct subtypes and progression patterns in colorectal cancer and systemic lupus erythematosus.

Legged locomotion enables robotic systems to traverse extremely challenging terrains. In many real-world scenarios, the terrain is not that difficult and these mixed terrain types introduce the need for flexible use of different walking strategies to achieve mission goals in a fast, reliable, and energy-efficient way. Six-legged robots have a high degree of flexibility and inherent stability that aids them in traversing even some of the most difficult terrains, such as collapsed buildings. However, their lack of fast walking gaits for easier surfaces is one reason why they are not commonly applied in these scenarios. This work presents LAURON VI, a six-legged robot platform for research on dynamic walking gaits as well as on autonomy for complex field missions. The robot's 18 series elastic joint actuators offer high-frequency interfaces for Cartesian impedance and pure torque control. We have designed, implemented, and compared three control approaches: kinematic-based, model-predictive, and reinforcement-learned controllers. The robot hardware and the different control approaches were extensively tested in a lab environment as well as on a Mars analog mission. The introduction of fast locomotion strategies for LAURON VI makes six-legged robots vastly more suitable for a wide range of real-world applications.

Incorporating Clinical Outcome Assessments (COAs) in oncology trials is essential for evaluating therapeutic benefits and balancing survival with quality of life and symptom management. Understanding COA adoption trends and gaps is especially critical in neuro-oncology due to the complex symptom profiles of neuro-oncology patients. We reviewed neuro-oncology trials in ClinicalTrials.gov, identifying COA instruments using the PROQOLID database. We analyzed trends in COA usage, associations with trial characteristics, and the impact of key regulatory and advocacy efforts using correlation and regression analyses. Of 1,874 adult interventional neuro-oncology trials with sufficient data, 85% were early-phase, 91% treatment-focused, primarily studying glioblastomas (61%), and unspecified gliomas (42%); 16% used COAs, with 127 distinct instruments reported; of these trials, 71% used Patient Reported Outcomes, 46% used Clinician-Reported Outcomes, with other categories of COAs used by fewer than 12% of the trials. COA use was more likely in later-phase trials (OR = 1.38, P <.00001), supportive care trials (OR > 1, P <.01), studies on novel versus FDA-approved interventions (OR = 1.72, P <.00001), single- versus multi-arm trials (OR = 1.62, P =.027), and randomized versus nonrandomized trials (OR = 3.28, P <.00001). COA incorporation increased over time (R = 0.77, P <.00001), especially in treatment-focused trials (R = 0.88, P <.00001) and across all COA types except composite measures. Our computational assessment highlights low but increasing COA adoption in neuro-oncology trials, mirroring broader oncology trends and reflecting a cumulative impact of regulatory and advocacy efforts. Priority areas for improvement include early-phase trials, FDA-approved interventions, and treatment-focused studies. Our findings emphasize the need for COA-specific reporting improvements and greater standardization in neuro-oncology research.

In recent years, Artificial Intelligence (AI) has developed advanced healthcare diagnostics by generating rapid growth and accurate disease detection. However, traditional approaches faced challenges like model interpretability, data security. To overcome these issues, hybrid approach Residual Network-50 with Federated Learning namely (ResNet-50-FL) is proposed for AI powered healthcare diagnostics. Initially, data is collected from ChestX-ray14 dataset which consists of over 100,000 frontal-view chest X-ray images annotated with 14 disease labels. Next, preprocessing is done by using image normalization and augmentation for medical images to improve the generalization of models. Then, feature selection is done by using Recursive Feature Elimination (RFE) were used to identify the most relevant predictors of patient outcomes. Finally, healthcare diagnostics is done by using proposed ResNet-50-FL to secure the patient raw data. From the results, the proposed ResNet-50-FL model attained better outcomes compared to existing ResNet-50 in terms of Accuracy, Sensitivity, Specificity, and Area Under the Receiver Operating Characteristic curve (AUC-ROC) by achieving 95.02%, 97.31%, 96.76% and 0.98% respectively.

Aerial vehicles are the primary enablers of urban low-altitude services. One of the key methods to achieve high data rate communication is the use of phased array narrow beams. However, equipping them can complicate the process of neighbor discovery, especially in the absence of clock synchronization among the nodes. This paper addresses the problem of asynchronous and prior information-agnostic Urban Aerial Ad-Hoc Networks (UAANs) and proposes a multi-Token-based network formation mechanism. By employing parallel token transmissions, the network formation duration is significantly reduced. Additionally, a neighbor cooperation mechanism is introduced during the network formation process, which improves the link number completeness through the Node Transition Phase. The simulation results show that, compared to D-SAND network formation strategy, when the number of nodes is 30, our strategy reduces the network formation time to 17.5% while ensuring network connectivity, at the cost of missing only 0.9% of the links. As the number of nodes grows, the advantages of our approach become increasingly evident.