Large language models (LLMs) often perform better when prompted to explain their reasoning, but it remains unclear how well such gains persist as reasoning depth increases. In this work, we propose a depth-aware evaluation framework alongside the performance results on two structured datasets: CLUTRR (kinship reasoning) and ProofWriter (logical entailment), comparing direct vs. reasoning (reasoning depth = number of inference steps required) prompts across five models. Reasoning gave small gains at shallow depths but quickly weakened and often reversed as tasks grew more complex. In ProofWriter, GPT-5 reached 90% accuracy at depth four in direct model, yet its reasoning accuracy fell below baseline after depth two. Smaller open-source models showed only unstable or negligible gains, underscoring that reasoning in LLMs remains brittle with increased depth.

Differential privacy (DP) is the leading standard for privacy protection, providing rigorous privacy guarantees for various data. However, its conventional approach of treating all records uniformly regarding privacy risk and using a non-adaptive privacy budget ($\epsilon$) often compromises data utility in subsequent analyses. This uniform treatment and fixed$\epsilon$can introduce significant perturbations, making the secondary use of shared data challenging. To overcome these limitations, we introduce a novel record-sensitivity-aware and Particle Swarm Optimization (PSO)-driven customised$\epsilon$-DP method for data perturbation. Our approach significantly enhances data utility without compromising privacy in data sharing by introducing three key optimisations to the traditional DP framework: First, we partition records into three sensitivity classes (high, medium, and low) based on the privacy risk. Second, we adopt a PSO mechanism to determine the optimal$\epsilon$for each partition, perturbing data with a variable$\epsilon$that considers sensitivity, rather than using a single, fixed$\epsilon$for the entire dataset. Finally, noise is injected by grouping attributes horizontally, rather than adding noise to each attribute independently, to prevent the generation of inconsistent values in the perturbed data. Detailed experiments on real benchmark and synthetic datasets demonstrate the superiority of our method in terms of utility and privacy across seven evaluation metrics, compared to the latest state-of-the-art$\epsilon$-DP methods.

Recently, data has ousted oil as the most economical resource in the world, but most companies are reluctant to share customer/user data in pure form and on a large scale due to privacy concerns. Many innovative technologies (e.g., federated learning, split learning) are employed to meet the growing demand for privacy preservation. Despite these technologies, acquiring personal data in order to optimize utility, and then sharing it on a large scale, is still very challenging. Thanks to the rapid development of artificial intelligence (AI), a relatively new and promising solution to resolve these challenges is to generate synthetic data (SD) by mirroring the original dataset’s properties. SD is a promising solution to address growing privacy demands as well as the utility/analytics requirements of many industry stakeholders. In this paper, we propose and implement an SD generation method from a real dataset containing both numerical and categorical attributes by using an improved conditional generative adversarial network (CGAN), and we quantify the feasibility of SD on technical and theoretical grounds. We provide a detailed analysis of SD in original and anonymized forms with the help of multiple use cases, whereas prior research simply assumed that privacy issues in SD are small because AI models do not overfit or SD has a poor connection with real data. We provide insights into the characteristics of SD (distributions, value frequencies, correlations, etc.) produced by the CGAN in relation to the real data. To the best of our knowledge, this is the pioneering work that provides an experiment-based analysis of the quality, privacy, and utility of SD in relation to a real benchmark dataset.

In this paper, we propose and implement a novel anonymization model, called data-centric$\ell$-diversity, to effectively safeguard the privacy of individuals with considerably enhanced utility in data publishing scenarios. Through experimental analysis of real-life datasets, we found that when the data quality is poor (e.g., distributions are uneven), most of the existing methods only anonymize some parts of the data (where distributions are balanced) and leave other parts unprocessed, which can lead to explicit privacy disclosures. Furthermore, they do not identify and repair problematic parts of the data before anonymization, and therefore, they are not secure from the threat of privacy breaches. To address these technical problems, in this paper, we implement an automated method that identifies vulnerabilities in the underlying data to be anonymized w.r.t. distribution, and that repairs them by injecting virtual samples of good quality. Later, we implement a data partitioning strategy that creates compact and diverse classes of size$k$, where$k$is the privacy parameter. Finally, only shallow generalization (or no generalization) is applied to each class to minimally generalize the data, whereas existing methods overly distort data by not improving the quality beforehand, which can lead to poor utility in data-driven services. We conducted detailed experiments on four datasets to justify the performance of our model in realistic scenarios, and achieved promising results from the perspectives of boosted accuracy, privacy preservation, data utility enrichment, and reduced computing overheads. Compared with baseline methods, our model enhanced privacy preservation by 36.56% on three different metrics, and data utility was augmented with 18.65% less information loss and 14.37% greater accuracy. Lastly, our model, on average, has shown a 26.13% reduction in time overheads on four datasets compared to the SOTA baseline methods.

Though anonymization mechanisms have made huge progress in fostering the secondary use of data, it is still very challenging to obtain adequate knowledge from anonymized data while preserving privacy. Most existing mechanisms anonymize entire sections of data and fail to maximally preserve the structure/values of real data. Consequently, the performance of those mechanisms and the output (i.e., the anonymized data) remain problematic in real-life scenarios due to the extensive and unneeded anonymization applied. To address these issues, we propose and implement a hybrid (differential privacy (DP) and$k$-anonymity) anonymization scheme that produces supreme-quality anonymized data that offers knowledge similar to real data without compromising privacy. Specifically, we implement a pair of algorithms that divide the dataset into privacy-violating and nonprivacy-violating partitions. Afterward, in a nonprivacy-violating partition, a relaxed privacy budget$\epsilon$is applied to numerical attributes, but most of the categorical attributes are retained (as is) for informative analysis. In privacy-violating partitions, fewer changes are applied to the data by using a reasonable value for$\epsilon$and by exploiting the diversity in sensitive information. Experiments are conducted on three real-life datasets to prove the feasibility of our scheme for futuristic AI applications. Compared with state-of-the-art (SOTA) methods, our scheme preserves 60.81% of the originality in the anonymized data. The privacy risks are reduced by 20.05%, and utility is enhanced by 54.01% and 15.33% based on information loss (IL) and accuracy metrics. Furthermore, the time overhead is 3.13$\times$lower than the SOTA methods.

To mitigate cybersecurity breaches, secure communication is crucial for the Internet of Things (IoT) environment. Data integrity is one of the most significant characteristics of security, which can be achieved by employing cryptographic hash functions. In view of the demand from IoT applications, the National Institute of Standards and Technology (NIST) initiated a standardization process for lightweight hash functions. This work presents field-programmable gate array (FPGA) implementations and carefully worked out optimizations of four Round-3 finalists in the NIST standardization process. A novel compact PHOTON-Beetle implementation is proposed wherein the underlying matrix multiplication is executed in serialized fashion to achieve a small hardware footprint. Sparkle implementations are carried out by implementing the ARX-box in serialized, parallelized, and hybrid approaches. For Ascon and Xoodyak, the proposed implementations compute certain permutation rounds in one clock cycle in order to explore the tradeoff between computation time and hardware area. As a result, this work achieves the smallest hardware footprint for PHOTON-Beetle consuming an area$3.4 \times $smaller than state-of-the-art implementations. Ascon and Xoodyak are implemented in a flexible manner that achieves throughput-to-area (TP/A) ratios$1.8 \times $and$3.9 \times $higher, respectively, compared to implementations found in the literature. In addition, we propose the first FPGA implementations for the Sparkle hash function. These efficient implementations provide guidelines for choosing a suitable architecture for applications in demand that can be employed in the IoT environment to achieve data integrity for various applications.

Abstract User attributes affect community (i.e., a group of people with some common properties/attributes) privacy in users’ data publishing because some attributes may expose multiple users’ identities and their associated sensitive information during published data analysis. User attributes such as gender, age, and race, may allow an adversary to form users’ communities based on their values, and launch sensitive information inference attack subsequently. As a result, explicit disclosure of private information of a specific users’ community can occur from the privacy preserved published data. Each item of user attributes impacts users’ community privacy differently, and some types of attributes are highly susceptible. More susceptible types of attributes enable multiple users’ unique identifications and sensitive information inferences more easily, and their presence in published data increases users’ community privacy risks. Most of the existing privacy models ignore the impact of susceptible attributes on user’s community privacy and they mainly focus on preserving the individual privacy in the released data. This paper presents a novel data anonymization algorithm that significantly improves users’ community privacy without sacrificing the guarantees on anonymous data utility in publishing data. The proposed algorithm quantifies the susceptibility of each attribute present in user’s dataset to effectively preserve users’ community privacy. Data generalization is performed adaptively by considering both user attributes’ susceptibility and entropy simultaneously. The proposed algorithm controls over-generalization of the data to enhance anonymous data utility for the legitimate information consumers. Due to the widespread applications of social networks (SNs), we focused on the SN users’ community privacy preserved and utility enhanced anonymous data publishing. The simulation results obtained from extensive experiments, and comparisons with the existing algorithms show the effectiveness of the proposed algorithm and verify the aforementioned claims.