Large Language Models (LLMs) are increasingly capable of supporting user decisions and performing various tasks autonomously in a wide range of professional domains, including the legal sector. In this work, we describe the general lifecycle of LLM-based systems and the architecture of a system designed to support legal experts in the process of generating maxims from judgments. We propose three application-specific performance metrics and a general-purpose assessment method suitable for the comparison of different systems in any content generation task. Finally, we discuss the results achieved by the system prototype that, according to a board of magistrates, outperforms human experts in the task of generating maxims from judgments.

High grade gliomas are malignant brain tumors with a median survival of under 16 months. Despite advances in treatment, which includes surgical resection followed by chemotherapy and radiation, this number has not increased much over the last decade. The success of immunotherapies in other cancers gives hope for such therapeutics in gliomas. This research utilized a publically available RNA sequencing data from 625 low-grade gliomas and 388 high-grade glioma samples from the Chinese Glioma Genome Atlas. Changes in gene expression between primary and recurrent tumors were analyzed as clinical trials are mostly conducted in the recurrent setting. The results indicated that the expression of several immune checkpoint molecules, like PD-1, PD-L1, CTLA4, and LAG3, did not change between primary and recurrent tumors. Conversely, the expression of OPN, B7-H3, TGF-Beta, and Galectin-3 significantly changed between primary and recurrent tumors and should be considered therapeutic for recurrent glioma patients. While the failure of several immunotherapies in clinical trials may be due to the low expression of molecules such as PD-1, there are currently trials underway for B7-H3 and TGF-Beta. In the future, clinical trials for gliomas should consider OPN and Galectin-3.

A document by Prasoon Garg. Click on the document to view its contents.

The efficiency of an energy receiver of a Wireless Power Transfer (WPT) system is intricately influenced by many factors such as frequency, power, topology, and the parameters of the rectifying diode. Developing an efficient and accurate model is critical for effective design. Unlike analytical models that rely on signal approximation and circuit simplification, the time-domain model can comprehensively incorporate all diode parameters and provide a complete view of the circuit with all time-domain signals available. However, to ensure accuracy and rapid convergence, the output low-pass section has to be handled carefully due to its DC voltage bias for the diodes and its long time constant. Furthermore, scaling the time-domain model for diverse rectenna topologies presents challenges. In this work, we propose a recursive time-domain model suitable for half-wave and any-stage voltage multiplier rectennas. The output low pass section is addressed by replacing it with an equivalent DC voltage source, thus avoiding the long settling time and ensuring accurate DC bias for the diodes. An analytical model for converting input voltage to DC calculates the equivalent source. Our proposed model is scalable for any number of multiplier stages and can accommodate the effect of matching networks. We validate the accuracy of calculated power conversion efficiency and rectifier input impedance by comparing them with results from Harmonic Balance simulations.

A document by Florian Reißner . Click on the document to view its contents.

Understanding the limitations of intelligent agents is crucial for optimizing their applications and mitigating potential risks. This paper explores the technical, functional, ethical, and practical limitations of agents, highlighting their impact on various domains and suggesting strategies for improvement.

Accurately detecting Parkinson's disease (PD) at an early stage is certainly indispensable for slowing down its progress and providing patients the possibility of accessing to disease-modifying therapy. An innovative deep-learning technique is proposed to predict and monitor the Parkinson's Disease (PD) through voice data and Unified Parkinson's Disease Rating Scale (UPDRS) score. The rise of an aging population over the world emphasizes the need to detect PD early, remotely and accurately. PD is a progressive neurological disorder that affects the Central Nervous System (CNS), leading to symptoms like tremors, stiffness, slow movements, balance and coordination difficulties, and speech disorders. Early detection and severity assessment of PD using Machine Learning is crucial. Speech recognition offers a new approach for diagnosis and monitoring of PD. This paper highlights Deep Learning models with acoustic features like jitter, shimmer, intensity and pitch for automatic detection and use of UPDRS score for severity assessment. Two datasets, one of them containing speech samples from PD patients and 1 healthy individuals and another containing voice data and UPDRS score, with on disk size of 5.3 MB and 1.01 MB respectively, are used in this proposed work. A comparison between the proposed Residual Neural Network (ResNet) and other ten machine learning and ensemble learning methods based on relatively small data including 64 healthy individuals and 188 early PD patients shows the superior detection performance of the designed model, which achieves the highest accuracy, 98 % on average and precision, 0.98 and F1-Score, 0.98.

The Internet of Things or IoT is the vast network of physically connected devices and the technology that allow them to communicated between these devices and the cloud, as well as between the devices themselves. IoT devices also allow to transmit, receive and exchange data with other device, applications and systems like WSN and RFID. IoT devices provide a lot of facilities to human that commonly use in home, transportation, education, healthcare and business. However, due to the weak security measures this paper will review the current state of IoT security which is highlighting the vulnerabilities that expose these devices to hacking, which can be used to launch attacks on other devices or networks. For example, weak password and insecure data transfer and storage. Therefore, the preventions knowledge needs to know like using blockchain technology, security analytics, API, PKI and authentication to detect and prevent attacks.

Network Function Virtualization contributes a lot to the networking environments. Many service providers gain benefits from the usage of NFV architecture. NFV works by virtualizing the works and functionality offered by individual hardware networking components, thus making it easily accessible via virtual form. Organizations can reduce the usage of hardware appliances because of NFV’s approach. There are a good number of related studies related to NFV which show people’s interest in this topic. Various research methodologies are used for this study such as researching, data collection, and summarizing. Next, the outstanding performance of NFV leads to providing many benefits such as cost efficiency, flexibility, and reduced energy consumption. However, despite the advantages, NFV also encounters several security challenges. This study will cover the issues of topology risks, complex layers, and malicious insiders. To overcome these challenges, this study suggests strong encryption and authentication, standardized interfaces and APIs, and Zero Trust security architecture.

Both the Restricted Hopfield Network (RHN) and Dense Associative Memory (DAM) are derived from the classical Hopfield Neural Network (HNN), but they improve upon HNN using different mechanisms. This paper presents a comparative analysis of RHN and DAM with respect to storage capacity, time complexity, training efficiency, and retrieval from incomplete or noisy patterns. The analysis reveals that RHN and DAM use different mechanisms for storing patterns. RHN, trained using the Subspace Rotation Algorithm (SRA), exhibits better time complexity than DAM, which is trained using the Energy-based Algorithm (EBA). Furthermore, in practice, RHN requires significantly less time than DAM to memorize the same number of patterns, as DAM needs more epochs to converge. The experimental results demonstrate that when storing 1,000 character patterns (825 bits), RHN with two hidden layers outperforms DAM in terms of retrieval from noisy and incomplete patterns. When memorizing 10, 50, and 100 patterns from both the MNIST and Fashion MNIST datasets, RHN shows a slight advantage over DAM in retrieval from incomplete and noisy patterns. However, the performance of both models degrades as the number of hidden nodes or stored patterns increases.  The link to the source code: https://github.com/Developer2046/rhn_dam.

An identity and data fiduciary model for source validation in social networks is needed, one that protects users' privacy and independence, while ensuring that data being transmitted is authentic and human-generated. Such a model will enable new types of reputation scoring services to emerge, utilizing data on the networks based on the pseudonymous user's posts and transaction history.

As machine learning continues to find applications, extensive research is currently underway across various domains. This study examines the current methodologies of machine learning being investigated to optimize semiconductor manufacturing processes. Our research involved searching the SPIE Digital Library, IEEE Xplore, and ArXiv databases, identifying 58 publications in the field of machine learning-based semiconductor process optimization. These investigations employ machine learning techniques such as feature extraction, feature selection, neural network architecture, and are analyzed using different algorithms. These models find applications in advanced process control, virtual metrology, and quality control, critical aspects in semiconductor manufacturing for enhancing throughput and reducing production costs. We categorize the papers based on the methods and applications employed, summarizing the primary findings. Furthermore, we discuss the general conclusions of several studies. Overall, the reviewed literature suggests that machine learning-based semiconductor manufacturing is rapidly gaining popularity and advancing at a swift pace.

The synergy of reinforcement learning (RL)-based energy management and vehicle-to-everything communication has been proved effective in boosting the fuel economy of connected plug-in hybrid electric vehicles (PHEVs). However, the intricate coupling of mechanical, electrical, thermal states and driving cycle results in a high-dimensional complex energy control problem for PHEVs, which is challenging to optimally solve within the same time scale. To this end, this study designs a multi-horizon reinforcement learning (MHRL)-based energy management of PHEVs, aware of the traffic preview from intelligent transportation systems to optimize the energy flow and thermal states as well as the transient dynamics of the powertrain. The proposed strategy features a novel state space representation, and solves the coordinated training among multiple sub-networks belonging to different control tasks in various time scales. Simulation and hardware-in-the-loop experiments are carried out based on a standard driving cycle and a real-world driving cycle with real-time traffic data demonstrate that the MHRL strategy improves fuel economy by 3.0%~7.9% compared to conventional RL-based energy management under various coolant temperature conditions and dynamic driving scenarios.

Grid operators in areas with lots of solar generation need solar power forecasts to support decision making. Most forecasts that are available are deterministic, and most of the ones that are probabilistic are for single plants, meaning don't have combined probabilities for aggregations of all the plants in a region. Additionally, few methods for producing probabilistic forecasts are easy to replicate or deploy for grid operators. This work presents a method for producing regional probabilistic forecasts from an existing deterministic forecast using open source tools and freely available data. The method is demonstrated using a deterministic forecast that is produced with similar tools and data. Reference code to replicate this work, which is freely available, is also introduced.

The Internet of Things (IoT) is a grand vision of connecting billions of devices with heterogeneous types and capabilities. Even though it is an attractive environment that could change the way we interact with the devices, the real-life and large-scale implementation of it is greatly impeded by the potential security risks that it is susceptible to. The promise is great but the security challenges are also daunting indeed. The IoT security can be addressed from different angles but one of the key issues is the access control model. This would eventually determine which device would be given access to the IoT systems and which would be denied access. In this work, we have done a systematic and thorough survey on the state-of-the-art access control models in IoT. We have covered conventional as well as the advanced access control models taking the crucial period of various studies in this particular area. A number of critical questions have been answered and key works have been summarized. Based on our investigation, some insights into the area, its requirements, scale, and future challenges and prospects have been discussed. We have also provided our rationale and motivation for this work and compared it with previous surveys, none of which has captured this security aspect of IoT with such a depth and detail.

A document by Zeynep Kumralbaş. Click on the document to view its contents.

Contemporary social media platforms like Twitter have become the portals of public conversation, connecting billions of users worldwide. However, these domains have also evolved into platforms for hate speech, offensive content, and misinformation. While machine learning (ML) surrounding natural language processing (NLP) has evolved to detect inappropriate tweets on social media, there has been a void of research examining the weaknesses of these techniques against adversarial attacks. In this paper, we aimed to address a significant research gap between LLM model vulnerability against adversarial attacks and improve classification accuracy by successfully integrating the PoS tag filters. The research presented an in-depth literature study surrounding adversarial attacks within the context of NLP. These findings indicate that LLMs like DistilBERT, GPT2, and GPT Neo models are highly vulnerable to adversarial perturbation. However, incorporating PoS tag filters helped reduce performance deterioration in Roberta and XLM models. By highlighting specific weaknesses of LLMs and presenting a viable method to exploit these weaknesses more effectively, this study paves the way for future research aimed at fortifying these models against adversarial threats.

We study content update and delivery in an edge caching system consisting of a sensor, a set of users, and an aggregator. The users have an aversion to stale contents, quantified as the contents' version age cost, and they occasionally request the aggregator for updated contents. The aggregator fetches the contents from the sensor and serves the users on demand. We study optimal content fetching and serving problem, aiming to minimize the time-averaged content fetching, transmission and age costs. The problem lends itself to Markov decision problem framework albeit with a high dimensional state space due to different content ages at different users. We first consider a single user problem and derive the optimal policy. We then propose a heuristic solution to the multiuser problem, which is obtained by combining solutions to certain fictitious, augmented versions of single user problems. Finally, we numerically illustrate the derived content update and delivery policies and their properties.