Tennessee Cancer Specialists

Two-dimensional (2D) transition metal dichalcogenides (TMDs) have emerged as promising materials for electronics, energy storage, and solar energy systems. Chemical vapor deposition (CVD) offers a scalable bottom-up route for synthesizing high-quality TMD films; however, optimization of operating conditions is challenging due to strongly coupled mass, momentum, and heat transfer phenomena. To address these complexities, a detailed CFD model incorporating transport equations and reaction kinetics is developed and validated by using published experimental data. A systematic design of experiments is then applied to generate a data set linking the operating temperature, pressure, and reactant flow rates to key film characteristics, including the deposition rate and uniformity. Surrogate machine learning models have been developed to reduce the computational cost associated with CFD simulations. Among several algorithms evaluated, Extreme Gradient Boosting (XGBoost) provides the highest predictive accuracy for MoS2 film growth behavior. Model interpretability is further enhanced using SHapley Additive exPlanations (SHAP), which reveal how reactor variables interact and contribute to deposition outcomes. Finally, the XGBoost model is integrated with the nondominated sorting genetic algorithm II (NSGA-II) to simultaneously maximize film thickness and uniformity in a cold-wall horizontal CVD reactor. The resulting optimal operating windows are visualized through parallel coordinate plots, providing practical guidelines for improving MoS2 film quality and enhancing CVD process efficiency.

Orthogonal Time Frequency Space (OTFS) modulation is a promising candidate for Integrated Sensing and Communication (ISAC) in high-mobility vehicular scenarios. It offers reliable performance under challenging channel conditions. In this work, a vehi

Large Language Models (LLMs) have become integral to modern business operations, especially for tasks involving reasoning over large datasets. One prominent application of LLMs is in chatbot systems, where customers provide natural language queries, often complex in nature, requiring decomposition to retrieve relevant information from various data sources. These queries may span structured databases, unstructured data, or public information from the internet, making efficient data retrieval and reasoning vital for real-time, accurate responses. In this paper, we propose two cost-efficient ''Query Orchestration'' approaches (Context Latent and Context Acute) to address these challenges. By leveraging graph-based retrieval-augmented generation (RAG) techniques for vector search, we optimize data retrieval while minimizing reliance on LLMs for reasoning to reduce costs. Our approach is validated through experiments on a banking use case, where we demonstrate its effectiveness in providing high-quality

In many futuristic communication systems, integrated sensing and communication (ISAC) is emerging as a crucial use case, with wide applicability in vehicle-to-everything (V2X) scenarios. As orthogonal time frequency space (OTFS) is being investigated as a possible waveform for high-velocity scenarios, in this work we address the challenging fractional delay-Doppler (DD) bin shifts estimation by considering an OTFS-based multistatic ISAC system. We make the problem pilot agnostic and propose a novel two-dimensional delay-Doppler (2D-DD) virtual array incorporating the bin shifts as phase information. To jointly estimate the fractional DD bin shifts, two algorithms, DD-OMP and DD-ESPRIT, are proposed using such a 2D-DD virtual array, based on orthogonal matching pursuit (OMP) algorithm and estimation of signal parameters using rota-tional invariance (ESPRIT), respectively. The proposed algorithms are further analyzed to determine: a) maximum number of targets that can be identified using DD-OMP and DD-ESPRIT; b) Cramer-Rao lower bound (CRLB); and c) computational complexity. Simulation results underscore the effectiveness of the proposed algorithms to achieve improved target parameter estimation and communication link performance. Being super-resolution schemes, this improvement in performance is reflected even at lower bandwidths and is independent of the pilot structure, thus making them attractive for ISAC systems.

We consider a periodic double auction (PDA) wherein the main participants are wholesale suppliers and brokers representing retailers. The suppliers are represented by a composite supply curve and the brokers are represented by individual bids. Additionally, the brokers can also participate in small-scale selling by placing individual asks; hence, they act as prosumers. Specifically, in a PDA, the prosumers who are net buyers have multiple opportunities to buy or sell multiple units of a commodity with the aim of minimising the cost of buying across multiple rounds of the PDA. Formulating optimal bidding strategies for such a PDA setting involves planning across current and future rounds while taking into account the bidding strategies of other agents. In this work, we propose Markov perfect Nash equilibrium (MPNE) policies for a setup where multiple prosumers with knowledge of the composite supply curve compete to procure commodities. Thereafter, the MPNE policies are used to develop an algorithm called MPNE-BBS for the case wherein the prosumers need to re-construct an approximate composite supply curve using past auction information. The efficacy of the proposed algorithm is demonstrated on the PowerTAC wholesale market simulator against several baselines and state-of-the-art bidding policies.

Deep Neural Network (DNN)-based recommendation models (RMs) are widely adopted in enterprise applications to suggest products, videos, tweets, and posts to users. These models heavily rely on embedding tables, which contain latent embedding vectors and are accessed to evaluate the probability of user interactions. Recently published, Hetero-Rec framework leverages the embedding access history to allocate storage partitions in heterogeneous memory architecture, aiming to maximize the chance of having frequently accessed embeddings available in faster memory. In this work, we extend the study of Hetero-Rec to larger embedding tables (up to 350), low-hot embedding tables, and evaluate the end-to-end speedups. Consequently, we present Hetro-Rec++ with heuristic based pre-optimizer and advanced formulation of optimizer’s cost function. Further, we demonstrate its effectiveness in reducing the embedding’s average fetch latency and hence, improving the inference latency for RMs deployed on Field Programmable Gate Arrays (FPGAs) with heterogeneous memory architectures.

Many complex workflows consist of multiple tasks represented as a directed acyclic graph (DAG). Optimal deployment of such workflows on a cloud using multiple services requires a judicious selection of compute and storage services to minimize the makespan and the cost of deployment. Multi-cloud deployment is emerging as a preferred choice for the deployment of complex workflows for price competitiveness and freedom from vendor lock-in. However, finding an optimal mapping scheme for heterogeneous tasks of a workflow in a multi-cloud environment is a challenge. Furthermore, each participating cloud service provider (CSP) has a unique cost model and maximum deliverable performance. This makes exploration of the optimal configuration of the chosen service daunting. Many algorithms, frameworks, and tools have been proposed to schedule complex workflows on cloud using virtual machines (VMs) available as Infrastructure-as-a-Service (IaaS). However, the use of scalable and cost-effective serverless platforms offered as Function-as-a-Service (FaaS) is still in its infancy.In this work, we use particle swarm optimization (PSO) for mapping complex workflows to cloud services such as computing, and storage in a multi-cloud scenario. We map complex workflows to the serverless platforms and storage services from popular cloud vendors namely Amazon Web Services (AWS), Azure (AZR), and Google Cloud Platform (GCP). The experimental evaluation shows that our approach results in up to 61% improvement in makespan and 51% improvement in the cost of workflow deployment as compared to naive and intuition-based mapping in a cloud.

Financial audits establish trust in the governance and processes in an organization, but they are time-consuming and knowledge intensive. To increase the effectiveness of financial audit, we address the task of generating audit suggestions that can help auditors to focus their investigations. Specifically, we present NLP techniques to extract hidden knowledge from a corpus of past financial audit reports of many companies, and use it for generating audit suggestions. The extracted knowledge consists of a set of automatically identified sentences containing adverse remarks, the financial variables mentioned in each sentence and automatically assigned XBRL categories for them, since XBRL is a standardized taxonomy in the financial domain. In the absence of suitable labeled data, we adopted a weak supervision approach. We designed a set of high precision linguistic rules to identify adverse remark sentences, created automatically labeled training data using them, and trained BERT-based and other classifiers to identify such sentences. We next presented novel techniques (which are either unsupervised or zero-shot) to assign zero, one, or more XBRL categories to any given adverse remark sentence. We evaluated the proposed approaches, on a large corpus of real-life financial statements and audit reports, against competent baselines. Given a company’s financial statements (already identified as suspicious), and given a subset of financial variables in them that contribute to suspiciousness, we match these with the extracted knowledge base and identify aligned adverse remarks that help the auditor in focusing on specific directions for further investigations.

Tracking arm motion in real-time with wearables like smartwatches has proved to be a challenging task. Ex-isting approaches either require additional sensing modality affecting ubiquity, and/or are not efficient enough for real-time tracking, especially on low-power embedded devices with limited computing capabilities. In addition, tracking air-writing trajectories while the user is on the move, e.g. walking, adds up to the challenges. ‘NNTrak’ is our attempt to address these issues focusing on practicality, with current work exploring air-writing scenarios. Inertial data from a smartwatch is used to train a computationally inexpensive Neural Network. It learns air-writing dynamics by modeling multi-direction straight line patterns of arm movement, providing 2-D directional inferences while air-writing irrespective of plane of writing around the user. This effects a real-time trajectory output. Multiple air-writing use-cases are tested to establish proof of concept, aiming at a broader 3D tracking in future. NNTrak's air-writing trajectories are compared with Kinect V2 trajectories. Independent-user evaluation is performed for use-cases like contact-less mouse pointer control with air-writing achieving a throughput of 2.4 bits/s and an average movement time of 1.7s for “medium” difficulty ISO task. A training-free gesture recognizer is designed using NNTrak and shape matching, achieving gesture detection accuracy of 93.6 % in a free-living environment. All experiments were performed on a computationally constrained Raspberry Pi 3B. As a first in domain, we also demonstrate NNTrak's capability in challenging scenarios like air-writing while the user is walking on stairs, doing light jogging etc. with interesting results.