Repository UNSAP (Universitas Sebelas April)
Not a member yet
1145 research outputs found
Sort by
Beyond fine-tuning: LoRA modules boost near-OOD detection and LLM security
No full textUnder resource constraints, LLMs are usually fine- tuned with additional knowledge using Parameter Efficient Fine-Tuning (PEFT), using Low-Rank Adaptation (LoRA) modules. In fact, LoRA injects a new set of small trainable matrices to adapt an LLM to a new task, while keeping the latter frozen. At deployment, LoRA weights are subsequently merged with the LLM weights to speed up inference. In this work, we show how to exploit the unmerged LoRA’s embedding to boost the performance of Out-Of-Distribution (OOD) detectors, especially in the more challenging near- OOD scenarios. Accordingly, we demonstrate how improving OOD detection also helps in characterizing wrong predictions in downstream tasks, a fundamental aspect to improve the reliability of LLMs. Moreover, we will present a use-case in which the sensitivity of LoRA modules and OOD detection are employed together to alert stakeholders about new model updates. This scenario is particularly important when LLMs are out-sourced. Indeed, test functions should be applied as soon as the model changes the version in order to adapt prompts in the downstream applications. In order to validate our method, we performed tests on Multiple Choice Question Answering datasets, by focusing on the medical domain as a fine-tuning task. Our results motivate the use of LoRA modules even after deployment, since they provide strong features for OOD detection for fine-tuning tasks and can be employed to improve the security of LLMs.LTS2Extended Abstrac
Brief Note on Secured P-Grid (Version-1)
No full textThis document talks briefly about making the current P- Grid implementation secure. We realize the following two requirements which we feel are essential for making communication on P-Grid the most secure one.LSI
Second Frcsyn-ongoing: Winning Solutions and Post-challenge Analysis to Improve Face Recognition with Synthetic Data
No full textSynthetic data is gaining increasing popularity for face recognition technologies, mainly due to the privacy concerns and challenges associated with obtaining real data, including diverse scenarios, quality, and demographic groups, among others. It also offers some advantages over real data, such as the large amount of data that can be generated or the ability to customize it to adapt to specific problem-solving needs. To effectively use such data, face recognition models should also be specifically designed to exploit synthetic data to its fullest potential. In order to promote the proposal of novel Generative AI methods and synthetic data, and investigate the application of synthetic data to better train face recognition systems, we introduce the 2nd FRCSyn-onGoing challenge, based on the 2nd Face Recognition Challenge in the Era of Synthetic Data (FRCSyn), originally launched at CVPR 2024. This is an ongoing challenge that provides researchers with an accessible platform to benchmark (i) the proposal of novel Generative AI methods and synthetic data, and (ii) novel face recognition systems that are specifically proposed to take advantage of synthetic data. We focus on exploring the use of synthetic data both individually and in combination with real data to solve current challenges in face recognition such as demographic bias, domain adaptation, and performance constraints in demanding situations, such as age disparities between training and testing, changes in the pose, or occlusions. Very interesting findings are obtained in this second edition, including a direct comparison with the first one, in which synthetic databases were restricted to DCFace and GANDiffFace.LASECLIDIA
Abstract 4588: Cell states and resistance in colorectal cancer: What can we learn from patient-derived organoids?
No full textColorectal cancer (CRC) is among the three most commonly diagnosed cancers and one of the major causes of cancer-related morbidity and mortality. Despite the discovery of several predictive biomarkers, the 5-year survival rate for metastatic CRC is 15%. A hallmark of CRC is inter- and intra-patient heterogeneity. Moreover, chemo-radiation and targeted therapies often result in producing resistant clonal populations that are unresponsive to the ongoing treatment. Revealing a comprehensive landscape of primary and metastatic CRC cell states and their evolution during chemotherapy treatment provides the potential for tackling this problem. Patient-derived organoids are a valuable research tool to study the biology of the disease as well as for drug screening in personalized medicine. To that end, we have collected a biobank of 90 PDOs from primary and metastatic CRC patients, as well as autologous cancer-associated fibroblasts (CAFs) and tumor infiltrating lymphocytes (TILs). We have performed whole-exome sequencing (WES), scRNA-seq and bulk RNA-seq of original tumors and PDOs, with and without chemotherapy treatment. We have discovered specific and common transcriptional cell states within PDOs and tumors, as well as the multiple responses to treatment confirming high heterogeneity of CRC. Combination of these results with clinical data can predict new biomarkers. Maxim Norkin, Pablo Hernandez Lopez, Cinzia Esposito, George Ramzy, Mireia Andreu Carbo, Paloma Ordóñez-Morán, Patrycja Nowak-Sliwinska, Salvatore Piscuoglio, Krisztian Homicsko, Joerg Huelsken. Cell states and resistance in colorectal cancer: What can we learn from patient-derived organoids [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2025; Part 1 (Regular Abstracts); 2025 Apr 25-30; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2025;85(8_Suppl_1):Abstract nr 4588.UPHUELSKE
Assessment of tire-derived additives and their metabolites into fruit, root and leafy vegetables and evaluation of dietary intake in Swiss adults
No full textTire wear particles, released at an estimated 6 million tons annually worldwide, introduce various chemical substances into agricultural environments through atmospheric deposition, road runoff, and reclaimed wastewater. These tire-derived compounds are known to impact ecosystem health. This study investigates the transfer of such additives and their metabolites into vegetables, assessing human dietary intake. Using UPLC-MS/MS, eleven tire-related compounds were analyzed in 100 vegetable samples from nine Swiss retailers, including leafy (lettuce, cabbage, spinach), root (onion, potato, carrot), and fruit (tomato, bell pepper, zucchini, pumpkin) vegetables. Contamination was detected in all vegetable varieties. 31 % of the 100 samples contained benzothiazole (BTH), 1,3-diphenylguanidine (DPG), 6-PPD, or 1,3-dicyclohexylurea (DCU) at levels exceeding the limit of quantification (LOQ) whereas blank values remained below LOD. DPG was most frequently detected (18 %, n = 100), followed by 6-PPD (15 %, n = 100), DCU (10 %, n = 100), and BTH (3 %, n = 100). Spinach comprised 78 % of DPG-positive leafy samples. Daily intakes of 6-PPDQ, DCU, 6-PPD, and DPG from vegetables were estimated at 0–18.7, 0–57.7, 0–42.3, and 0–42.4 ng/person/day, respectively. While current toxicological data suggest no immediate health concerns, significant knowledge gaps remain regarding long-term toxicity. This study offers critical insights into the presence of tire-derived substances in agriculture and underscores the need for further research to better assess environmental and human health risks.GR-CE
Comparison of Different Precision Pseudo Resistor Realizations in the DC-Feedback of Capacitive Transimpedance Amplifiers
No full textIn this paper, three different pseudo resistor (PR) realizations are compared regarding their noise and linearity performance within the DC feedback loop of an integrator-differentiator transimpedance amplifier (I-D-TIA). Thanks to its intrinsic shot noise suppression, the multi-element pseudo resistor (MEPR) shows the best SNDR for small AC signals around a large DC current. In contrast, for large AC signals around comparably small DC currents, the transconductor based PR introduced by Ferrari et al. in [1] displays a rail-to-rail signal swing capability and a small THD of 0.4% resulting in the best achievable SNDR value. All simulations have been performed on transistor level in a 180nm SOI CMOS technology with a 1.8V supply.LBN
First observation of CP violation and improved measurement of the branching fraction and polarization of B-0 -> D*(+) D*(-) decays
No full textWe report the measurement of the branching fraction, the polarization, and the parameters of the time-dependent CP violation in B-0 -> D*(+) D*(-) decays using a data sample of 772 x 10(6)B (B) over bar pairs, collected at the Upsilon(4S) resonance with the Belle detector at the KEKB asymmetric-energy e(+)e(-) collider. We obtain a branching fraction of B = (7.82 +/- 0.38 +/- 0.63) x 10(-4), a CP-odd fraction of R-perpendicular to = 0.138 +/- 0.024 +/- 0.006 and, additionally, a fraction of the longitudinal component in the transversity base of R-0 = 0.624 +/- 0.029 +/- 0.011. The measured values of the parameters of the CP violation are SD*+ D*- = 0.79 +/- 0.13 +/- 0.03 and A(D*+ D*-) 0.15 +/- 0.08 +/- 0.04.LPH