The DOST Advanced Science and Technology Institute is a research and development organization based in the Quezon City, Philippines. It is one of the research and development institutes of the Department of Science and Technology of the Philippine government.
The development of enhanced performance of biobased composite films with hydrophobic, chemical resistance, thermal and electrical properties.
Recent advances in electron microscopy and computer vision now allow the reconstruction of complete wiring diagrams, or connectomes, of animal brains. This creates an urgent need for methods that can automatically identify neuronal cell types directly from these large connectivity datasets. Here we show that synaptic connectivity alone can be used to assign neurons to cell types with high accuracy. We introduce NTAC (Neuronal Type Assignment from Connectivity), which groups neurons based only on connectivity. NTAC has two forms: a semi-supervised one that leverages a small fraction of labeled neurons to infer the types of all others, and an unsupervised one that requires no labels at all. Applied to multiple state-of-the-art fruit fly brain connectomes, NTAC achieves high accuracy within only minutes on a laptop, demonstrating that connectivity provides a powerful and scalable basis for classifying neuronal cell types across the brain. In this study, the authors develop NTAC, Neuronal Type Assignment from Connectivity, using synaptic connectivity alone to identify cell types with high accuracy within minutes on a standard CPU.
Abstractive summarization with large pre-trained Transformer models struggles to reliably control stylistic attributes (e.g., “punchy” vs. “neutral” tone) without sacrificing content fidelity, especially under parameter-efficient adaptation. This work systematically examines Low-Rank Adaptation (LoRA) on PEGASUS-Large for style-conditioned headline generation using special control tokens and an ablation over eight configurations that vary LoRA rank ($\mathbf{r} \in\{8,16,32,64\}$) and target modules (Attention-Only vs. FFN-Expanded). Experiments reveal a phenomenon we term Stylistic Collapse, where the model's strong extractive bias overwhelms the parameter-efficient style signal. Across all configurations, the Identical Output Rate remains high (above 62 %), with no consistent improvement from increased LoRA capacity, even though factual consistency (entailment $>0.85$) and content fidelity (ROUGE-1 $\approx 0.52$ for neutral) remain strong. These results suggest that, in our PEGASUS-Large setting and dataset, standard parameter-efficient fine-tuning may be insufficient for enforcing subtle stylistic control, and motivates future exploration of objectives (e.g., contrastive losses or reinforcement learning) that explicitly reward stylistic divergence from the base model's inductive bias.
This study aimed to identify high-yielding and stable bread wheat genotypes through genotype × environment interaction analysis using GGE biplot methodology. A total of 160 genotypes were evaluated during the 2023 and 2024 cropping seasons at Dabat and Adet in northwestern Ethiopia using an alpha lattice design with two replications. Significant effects of genotype, environment, and their interactions (p ≤ 0.0001) were observed across all eleven agronomic traits studied. The analysis revealed that genotypes G10, G36, G52, G81, and G28 consistently combined superior grain yield with high stability across environments. Among the test sites, Dabat 2024 emerged as an ideal environment for evaluating grain yield performance. Environmental clustering further delineated two distinct mega-environments, providing valuable insights for breeders in selecting genotypes with either specific or broad adaptation. These findings highlight the utility of GGE biplot analysis in guiding wheat breeding programs toward improved yield stability and targeted genotype deployment.
We investigate the coherent-elastic neutrino-nucleus scattering ( CEν NS ) in the SU(3)_L and generic U(1) gauge extension of standard model. Using data given by COHERENT experiment, we can obtain the lower bound for the new neutral gauge boson Z' . By evaluating Δχ ^2 , we show that in 331 models, m_Z'≥ 1.7 TeV for the CsI detector and m_Z'≥ 2.3 TeV for the liquid Argon detector at 90 m_Z'/g_Z'≥ 2.5 TeV. Our results for SU(3)_L model are consistent with electroweak and dark matter constraints on the Z' boson mass indicating that low-energy high-intensity measurements can provide a valuable probe complementary to high-energy collider searches at LHC.
