Digital Commons Network^™

Estate Planning: Charitable Transfers Of Life Insurance, Thomas H. Varner, Edward S. Schlesinger

Tax Adviser

No abstract provided.

Constructive Sales Price Rules For Excise Taxes Broadened By Tax Reform Act, Thomas H. Varner

Tax Adviser

No abstract provided.

Sec. 531 Tax—Working Capital Needs And The Operating Cycle, Mario P. Borini, Stanley L. Malaga, John P. Sullivan

Tax Adviser

No abstract provided.

Tax Planning-A Service Or A Crime?, K. S. Carmichael

Tax Adviser

No abstract provided.

Private Foundations And Tax Reform: Composition And Concepts, William J. Lehrfeld

Tax Adviser

No abstract provided.

Exploring Conditioning Strategies For Film-Based Feature Modulation In Saliency Ranking, Elena Roussanova, Anil Raut, Zaima Zarnaz, Seungbae Kim

Bellini College REU Symposium (BCREUS)

We present a parameter-efficient FiLM-based framework for visual saliency ranking that operates on a frozen ResNet backbone. Instead of fine-tuning convolutional weights, lightweight conditioning embeddings modulate four intermediate backbone stages through Feature-wise Linear Modulation (FiLM), enabling controlled scaling and shifting of both channel and spatial feature responses. The resulting multi-scale FiLM-modulated features are fused to produce instance-level saliency rank predictions. Our study shows that conditioning-based modulation can effectively reshape backbone representations while preserving parameter efficiency, providing a flexible alternative to full backbone tuning for fine-grained saliency reasoning.

Biologically Motivated Algorithms And Backpropagation, Yusra Rasool, Tasnia Mahsin Simin, Shion Matsumoto, Ankur Mali

Bellini College REU Symposium (BCREUS)

Backpropagation (BP) relies on biologically implausible mechanisms like global error transport. This project investigates two alternatives: Error-Driven Local Representation Alignment (LRA), which utilizes local error loops, and Predictive Forward-Forward (PFF), which employs forward-forward algorithm and "goodness" optimization to eliminate backward passes. We quantify the trade-off between biological plausibility and efficiency by measuring the convergence lag inherent to localized updates. Theoretically, we analyze how PFF and LRA resolve the weight transport problem and enable local credit assignment, positioning these frameworks as viable candidates for energy-efficient "mortal computation" on neuromorphic hardware.

Fine-Tuning Open Source Llms For Generating Pedagogically Aligned Practice Questions In University Level Algorithms Course, Anne Utegen, Tran Ho, Oguzhan Topsakal

Bellini College REU Symposium (BCREUS)

This project focuses on fine-tuning the open-source Gemma 3 model using Low-Rank Adaptation (LoRA) to generate pedagogically aligned practice questions for an Analysis of Algorithms course. The authors created a dataset consisting of 300 question-hint-answer conversations with progressive hints that would scaffold student reasoning. Gemma 3 was fine-tuned and evaluated against a prompt-only baseline using an LLM-as-a-judge framework (Gemini 2.5 Pro) across 187 question pairs. The results show that the fine-tuned model improved significantly in avoiding premature solution reveals, difficulty appropriateness, and pedagogical alignment. However, the model experienced decreases in hint quality, clarity, and correctness, which is hypothesized to be …

The “Surprisal” Between Contrastive Representation Learning And The Free Energy Principle, Raul Castillo, Abdul-Malik Zekri, Benjamin Prada, Ankur Mali

Bellini College REU Symposium (BCREUS)

The ultimate goal of machine learning is to build models that generalize well. different machine learning frameworks have different ways of achieving this goal. I-Con builds outputs that group like and separate unlike data points, building strong groupings at the cost of robustness against noisy/mislabeled data. The Free Energy Principle (FEP) builds outputs within a larger probability density, being able to account for noisy/mislabeled data at the cost of structure. We posit that I-Con and FEP can be formally related as Minimum Description Length (MDL) codelengths and as limit cases of Alpha-divergences. Rigorously proving this connection would provide new perspectives …

Alphafold3 And Intrinsically Disordered Proteins: Reliable Monomer Prediction, Unpredictable Multimer Performance, Tuan Minh Dao, Sebastien Ghent, Vladimir N. Uversky, Taseef Rahman

Bellini College REU Symposium (BCREUS)

AlphaFold3 represents a major advance in protein structure prediction, yet its performance on intrinsically disordered proteins remains uncharacterized. We present the first systematic evaluation of AF3 on disordered systems, revealing a striking dichotomy. For monomers, AF3's pLDDT scores reliably predict disorder (MCC: 0.693), matching AlphaFold2 and rivaling dedicated predictors. This consistency across fundamentally di ̄erent architectures confirms that disorder prediction emerges from training data, not model design. For multimers, the picture grows complex. Despite comparable aggregate performance (mean DockQ: 0.563 vs 0.571), AF3 and AF2 achieve these results through fundamentally di ̄erent mechanisms.

Conventional structural features explain 58% of AF2's …

Last-Mile Autonomous Delivery Robot, Lucas Araujo Bianco, Clement Joseph, Chance Hamilton, Alfredo Weitzenfeld

Bellini College REU Symposium (BCREUS)

Autonomous sidewalk navigation on campuses remains a significant challenge due to irregular ge- ometries, varying surface textures, and dynamic obstacles found in real-world settings. This project develops a ROS2-based system that enables the robot to navigate locally on the campus sidewalk and globally between campus buildings. We employ FastSCNN, a Semantic Convolutional Net- work trained on the Cityscapes dataset, allowing the robot to perform pixel-level classification of sidewalk and non-sidewalk in real time navigation. Applying transfer learning to a small, locally labeled dataset of our campus improved sidewalk segmentation, effectively adapting the pretrained model to our campus environment.

To translate …

Reps: A Multimodal Recipe Recommendation System For The Misty Social Robot, Jasmine Kaur Kohli, Alissa Josy, Jingjing Li, Zhao Han

Bellini College REU Symposium (BCREUS)

Meal-planning technologies such as recipe apps and diet trackers are widely used but typically rely on text or static visuals and lack social engagement, emotional responsiveness, and conversational adaptability. Social robots, with expressive speech, gesture, and display capabilities, offer the potential to make everyday tasks like cooking more accessible and enjoyable. Recent work highlights the growing role of large language models (LLMs) in robotics and embodied AI systems. This project presents REPS (Recipe Embodied Partner System), a multimodal recipe recommendation system for the Misty social robot that integrates structured recipe retrieval, LLM-based reasoning, and multimodal robot output to create a …

Evaluating Mid-Air Fog-Screen Robot System, Hong Wang, Ngoc Bao Dinh, Waylon Erlandson, Jacob Rauscher, Mario Sinclair, Nicolas Echeverria, Abrar Zahin, Zhao Han

Bellini College REU Symposium (BCREUS)

This project investigates how communication interfaces can bridge the gap between humans and robots in hazardous reconnaissance environments. The study explores a communication framework designed to overcome the limitations of headset-based solutions such as poor scalability, and projector-based spatial AR that rely on vertical surfaces often absent in open or unstructured environments. Through the development of a signaling system, the project aims to provide clear transmission of mission-critical safety information without reliance on headsets, network connectivity, or close-proximity displays. Previous results suggest that this approach improves information clarity for the operator compared to standard screen-based methods. The project advances this …

Selfbench-V And Unified Ppa Analysis Framework, Ghali Omar Boutaib, Marwan Abdelwahab, Tasnim Tabassum, Hao Zheng

Bellini College REU Symposium (BCREUS)

LLMs are a very hot topic in the EDA space at the moment. Most of the research efforts are focused on creating better models for Verilog code writing tasks, and datasets for training and fine tuning. A capitally important part of this research topic is the creation of benchmarks to evaluate the performance of these models once we create them. Over time and through research efforts, models get significantly better, and start clearing these benchmarks at very high rates. Too high rates. The creation of benchmarks is a very laborious task. It always involves elaborating novel sample design ideas, writing …

A Universal Evaluation Framework For Visual Noise In Contained Systems, Temirzhan Mukhambet, Duong Mai, Jacqueline Hausman, Yu Sun

Bellini College REU Symposium (BCREUS)

In many robotics or computer vision contained systems, regulating the quality of captured input data (e.g., point clouds or videos) is critical for preserving the integrity and accuracy of downstream tasks. In practice, this process requires mitigating various forms of noise specific to the domain (2D vs. 3D) and the intended application, such as object detection and scene understanding.

However, a unified framework addressing noise across a holistic system remains underexplored, as current surveys typically isolate specific modalities i.e., examine either 2D or 3D methods exclusively. The 2D domain covers two main areas: static image denoising, which prioritizes sensor noise …

Evaluation Of Deep Learning Approaches For Protein Function Prediction, Mohamed Elwaei, Isfahan Juboraj, Osama Abdeljabbar

Bellini College REU Symposium (BCREUS)

Protein function prediction is of high importance when it comes to applications such as bioinformatics, of which drug discovery and disease research, in particular, are key components. Experimental methods for determining protein function are accurate but typically cumbersome and expensive, and this is why computational approaches are needed for this. This study specifically explores, through a research-based approach, different deep learning methods to predict protein functions using amino acid sequences, structural information, and protein interaction data. The proposed framework combines Convolutional Neural Networks to capture local sequence patterns with Graph Neural Networks to model global structural relationships and protein interactions. …

Bloom: A Low-Cost, Open-Source, Companion Social Robot, Deekshita Senthil Kumar, Sebastian Ramirez-Vallejo, Xiangfei Kong, Zhao Han

Bellini College REU Symposium (BCREUS)

The rapid growth of the aging population has increased rates of loneliness, depression, and cognitive decline among older adults, particularly those living alone or with Alzheimer's disease and related dementias (ADRD). Socially assistive robots such as Paro have demonstrated therapeutic and emotional benefits; however, existing systems remain costly and inaccessible. This project introduces Bloom, a low-cost, physically interactive social robot designed to provide emotional companionship and cognitive engagement.

Built on a Raspberry Pi 5 platform and inspired by the open-source Blossom robot, Bloom features a durable 3D-printed hard-shell design with motor-driven expressive movements that support tactile interaction. The system integrates …

Analyzing Human–Robot Collaboration Failures In State-Of-Art Generative Visuomotor Models, Grant Henderson, Miguel Mateo Osorio Vela, Vidhi Mudaliar, Amanuel Ergogo, Zhao Han

Bellini College REU Symposium (BCREUS)

As generative visuomotor models become increasingly capable of representing large-scale robotic skills, they are beginning to transition from controlled laboratory environments into shared human-robot workspaces. While these models enable competent manipulation, their opaque, black-box decision-making processes introduce coordination challenges that can negatively impact safety, efficiency, and trust during collaboration. In safety-critical tasks such as medication dispensing, successful teamwork depends not only on physical capability but also on timing, predictability, and mutual awareness.

This project investigates how two state-of-the-art generative visuomotor policies, an Action-Chunking Transformer (ACT) and a Vision-Action Diffusion policy, perform in a real-world collaborative medication-dispensing task. We compare these …

Trauma-Informed Leaders: Shifting The Paradigm Of Professional Suicide Prevention, Janina Cich

Forensic Scholars Today

Trauma-informed Leaders: Shifting the Paradigm of Professional Suicide Prevention highlights how trauma-informed leadership can transform suicide prevention in law enforcement and other high-stress helping professions. Reflecting on current suicide data and first-responder research, the article outlines early warning signs, intervention strategies, and organizational wellness practices that drive agencies from reactive crisis response to proactive, culturally responsive care. Practical recommendations include leader-supported peer teams, sleep and micro-break interventions, and resilience-focused supports that foster psychological safety, reduce stigma, and normalize help‑seeking across the organization.

Forensic Neuroscience: A Brief Introduction, Jerrod Brown

Forensic Scholars Today

Forensic neuroscience is an emerging, multidisciplinary field that applies brain science to forensic mental health, criminal justice, and legal decision-making. This article summarizes how neuroscience can inform assessments, evaluations of criminal responsibility, case planning, treatment, and prevention for justice-involved populations, highlighting complementary disciplines and the need for ethical guidance, replication of findings, and specialized education before broad adoption in legal settings.

On-Device Measurement Of Impact Of Artificial Intelligence Tools On Productivity, Navid Kagalwalla, Tom Piacentini, Jacques Mouton

Defensive Publications Series

This disclosure describes techniques to measure the net productivity impact of AI-assisted workflows on user productivity. Metadata regarding user activity on a user device is obtained with user permission, using a private, on-device architecture. The data is analyzed on-device to determine whether an AI-assisted workflow is in process and the impact of the use of AI on productivity. In an example implementation, an Activity Sensing and Filtration Layer (ASFL) and a Cognitive Workflow Engine (CWE) are implemented and used to calculate the true net time saved using AI assistance. This metric takes into account the hidden cost of the user …

A Distance Based Embedding Method Using Nlp Methods To Identify Proximity To A Network, Shuhan Zhang, Anurag Tangri, Chinmay Narendra Sonar, Lin Meng, Can Liu, Chiranjeet Chetia

Defensive Publications Series

The present disclosure relates to the field of Artificial Intelligence (AI), in particular to AI methods and systems for analyzing transactional data using Natural Language Processing (NLP) techniques to identify latent relationships and proximity within networks. The disclosed system utilizes text fields within datasets to categorize input data into appropriate classifications. The input comprises a dataset containing text data that provides contextual information related to transactions. Feature embeddings are generated using a sentence-based Language Learning Model (LLM), which captures semantic meaning of sentences. The LLM converts input data of varying lengths into fixed-dimension embeddings. The sentence-based LLM is fine-tuned based …

Ai-Based Claims Assessment For Motor Vehicle Damage Detection, Anonymous

Defensive Publications Series

The insurance industry faces increasing pressure to optimize claims processing efficiency while minimizing claim leakage and fraud. Traditional rule-based and manual approaches are reaching their limits, particularly as market dynamics such as claims inflation and customer demand for faster, self-service experiences accelerate. To address these challenges, we present an AI-driven claims assessment framework that integrates computer vision and generative AI to support human interpretation of visual and textual evidence in vehicle damage claims.

The proposed system comprises two modules: Motor Damage Assessment for analyzing vehicle body damage and Glass Damage Assessment for evaluating windshield damage. The Motor Damage Assessment consists …

Llm-Augmented Dynamic Selective Blocking Of Service Calls For Application Control, Johannes Start, John Lunney

Defensive Publications Series

Software applications face critical challenges in rapidly and precisely mitigating emergent threats such as safety incidents or query-of-death attacks originating from or affecting backend calls. This disclosure describes techniques for dynamic, real-time, intelligent blocking or modification of harmful or problematic backend service call responses. The techniques combine traditional filtering with semantic analysis using a large language model (LLM). A control plane enables operators such as site reliability engineers (SREs) to define and deploy mitigation rules on-the-fly, without code changes or lengthy propagation delays. This enables rapid and precise response to emergent threats, such as safety incidents involving AI-generated content or …

Enhancing Hardware Scan Coverage By Selecting A Data Streaming Interface Unconnected To The Failure, Kaushal Purohit, Victor Wong

Defensive Publications Series

Hardware scanning helps debug problems related to internal components within a device by capturing hardware states as a stream of bits, known as a scan chain. However, employing on-chip or off-chip memory for storage of scan data requires that the path to/from the memory be excluded from the hardware scan, creating a critical gap in debugging. Moreover, wired debugging interfaces are typically unavailable or inaccessible on remote production devices, which makes it challenging to retrieve the scan data quickly. This disclosure describes techniques to optimize the streaming of hardware scan data from on-chip crash manager applications employed to detect hardware …

Closed Loop Correction Of Camera-To-Display Projection Using Piezoelectric Actuation, Na

Defensive Publications Series

Reconstructing a three-dimensional (3D) image of an object of interest requires high pixel accuracy between the optical plane and sensor images. While optical centers can be corrected for in the XY plane and along the roll axis with image cropping and rotation techniques, corrections based only on image processing are difficult in the yaw and pitch orientations. This disclosure describes techniques that leverage reference features in captured images calibrated with previously captured (yaw, pitch) data to correct, via trained machine learning models, the yaw and pitch of cameras. The camera orientations can be controlled by piezoelectric actuators.

Optimizing Data Transfer From Smart Glasses To Paired Device Based On Eye Tracking, Willy Horng Jean Cheung, Shenil Navin Dodhia, Gordon Wan

Defensive Publications Series

Smart glasses and other wearable devices that include cameras can capture the user’s environment as memories and can also answer questions or surface assistance proactively. However, smart glasses have limited compute and storage capabilities, and enabling continuous perception is therefore infeasible. A paired device such as a smartphone can enable smart glasses to operate in continuous perception mode and also provide compute power for artificial intelligence models. However, a key bottleneck is the data transfer between the smart glasses and the paired device, which is constrained by wireless data transfer rates. Even with aggressive data-compression schemes, the payload over the …

Dynamic Management Of Virtual Lanes And Buffer Zones For Autonomous Traffic, Sumukh Shevde, Kamran Mustafa, Abhijit Adsule

Defensive Publications Series

Static assignment of roadway lanes can result in inefficient traffic flow, where congestion may develop in one direction while opposing lanes remain underutilized. To address this, systems and methods for dynamic traffic management are described. A computational system, such as a central server or an in-vehicle computer, can treat a roadway as a single, partitionable surface, creating and modifying virtual lanes in near real-time. The number, width, and direction of these lanes can be adjusted based on current and predicted traffic demand. Autonomous or semi-autonomous vehicles can be assigned to these virtual lanes, and a dynamic virtual buffer zone may …

A Real-Time Computer Vision System For Procedural Verification In Manual Assembly, Frank Mao, Sun Nee Chan, George Liu

Defensive Publications Series

Manual assembly processes can be susceptible to error, potentially leading to incorrect component handling or procedural deviations. A system for real-time procedural verification can use computer vision to address this. An inspection node, which may comprise an imaging device (e.g., a camera) and a processing unit, can capture and analyze an operator's physical actions at a workstation. These actions may be compared against a predefined digital blueprint, such as a standard assembly model (SAM), that codifies a specified sequence of events. Upon detection of a deviation between the operator's actions and the SAM, the system can provide feedback to the …

Helping Commercial Trailer Fleet Operators Successfully Hook Up A Trailer & Allowing The Fleet Manager To Monitor Their Skill, Anonymous

Defensive Publications Series

Commercial users of a trailer feature where an operator may be required to hook up various trailers and various vehicles, some of which may have different specific connections or different orientation/requirements. The truck using its cameras and sound exciters can become a trailer validation coach/monitor for commercial and fleet vehicle operators, reducing risk and liability of incorrect connections.