447 research outputs found
Attribute reduction approaches for general relation decision systems
No full textThis paper proposes the concept of general relation decision systems and studies attribute reduction algorithms for relation decision systems, which are generalization of decision tables. In our relation decision systems, both condition and decision attribute sets consist of general binary relations. Novel attribute reduction algorithms for consistent and inconsistent relation decision systems are derived, respectively. A data set from the UCI machine learning databases is used in the empirical study, the experimental results verify the effectiveness of the proposed algorithms. The results unify the earlier attribute reduction algorithms for decision tables
Developing force field parameters for water interacting with graphene and graphene-like materials
No full textConfined water can have properties dramatically different from bulk water, and these properties can be used to develop unique functionality at the nanoscale. For example, fast water transport, rotation-translation coupling, and fast rotationalmotion have been observed in graphitic carbon-based nano structures, which enables various applications like energy storage and seawater desalination. The explosive studies on graphene have sparked new interests towards graphene-analogous materials including hexagonal boron nitride (hBN) and molybdenum disulfide (MoS2). Compared to graphene, the graphene-analogous materials possess non-zero bandgap, chemical inertness, and biological compatibility. The graphene-analogous materials are promising materials, complementary to graphene, for high-temperature, biomedical and nanofluidic applications. We would like to understand and optimize graphene and graphene-analogous materials in these applications.
The study of graphene and graphene-analogous materials at the atomic level requires accurate force field parameters to describe the water-surface interaction. We begin with benchmark quality first principles quantum Monte Carlo (QMC) calculations on the interaction energy between water and surface, which are used to validate random phase approximation (RPA) calculations. We then proceed with RPA to derive force field parameters, which are used to simulate properties like water contact angle on the surface, attaining a value within the experimental uncertainties. This work demonstrates that end-to-end multiscale modeling, starting at detailed many-body quantum mechanics, and ending with macroscopic properties, with the approximations controlled along the way, is feasible for these systems.Submission published under a 24 month embargo labeled 'U of I Access', the embargo will last until 2018-08-01The student, Yanbin Wu, accepted the attached license on 2016-07-01 at 11:25.The student, Yanbin Wu, submitted this Dissertation for approval on 2016-07-01 at 11:32.This Dissertation was approved for publication on 2016-07-05 at 09:40.DSpace SAF Submission Ingestion Package generated from Vireo submission #9740 on 2016-11-10 at 12:24:48Made available in DSpace on 2016-11-10T18:39:15Z (GMT). No. of bitstreams: 3
WU-DISSERTATION-2016.pdf: 3224743 bytes, checksum: 20bf91ec78585cb0900f4d33466d1dd4 (MD5)
LICENSE.txt: 4206 bytes, checksum: b74ce964236b5b29ccdd465d6a0ce916 (MD5)
PROQUEST_LICENSE.txt: 4552 bytes, checksum: a09a20759fd03a5655783048ec58163c (MD5)
Previous issue date: 2016-07-05Embargo set by: Seth Robbins for item 95447
Lift date: 2018-11-10T18:39:22Z
Reason: Author requested U of Illinois access only (OA after 2yrs) in Vireo ETD systemEmbargo set by: Seth Robbins for item 95447
Lift date: 2018-11-10T18:43:22Z
Reason: Author requested U of Illinois access only (OA after 2yrs) in Vireo ETD systemU of I Only Restriction Lifted for Item 95447 on 2018-11-11T10:15:28Z
Towards Superior Quantization for Large Language Models
No full textLarge Language Models (LLMs) have exhibited remarkable capabilities in tasks such as natural language comprehension, content generation, and knowledge retrieval. However, training and serving these models require substantial computational resources, posing a significant barrier to AI application development and research. To address these challenges, various model compression techniques have been explored, with quantization emerging as a key approach. Nonetheless, existing quantization methods predominantly apply uniform quantization configurations, failing to account for the varying quantization difficulty across different layers in billion-scale models. This results in a rigid memory-accuracy trade-off and leaves the potential for improving quantization accuracy through differentiated memory allocation largely unexplored. To bridge these research gaps, this thesis advances the study of LLM quantization with two key contributions. First, it introduces MXQ, a mixed-quantization method designed to provide a more flexible memory-accuracy balance. MXQ formulates a novel optimization approach to determine optimal layer-wise quantization parameters while enforcing overall memory constraints. Experimental results demonstrate that MXQ unlocks a broader spectrum of quantization configurations, simplifying the memory-accuracy trade-off while maintaining performance comparable to the baseline. Second, this thesis proposes SensiBoost and KurtBoost, two methods that enhance quantization accuracy by leveraging layer-sensitive features such as activation sensitivity and weight distribution kurtosis to identify critical layers. These approaches outperform existing baselines, achieving up to 9% lower perplexity with only a 2% increase in memory budget on Llama models
The thermal properties of AlGaAs/GaAs laser diode bars analyzed by the transient thermal technique
No full textGenomes of multicellular algal sisters to land plants illuminate signaling network evolution
No full textAbstract Zygnematophyceae are the algal sisters of land plants. Here we sequenced four genomes of filamentous Zygnematophyceae, including chromosome-scale assemblies for three strains of Zygnema circumcarinatum . We inferred traits in the ancestor of Zygnematophyceae and land plants that might have ushered in the conquest of land by plants: expanded genes for signaling cascades, environmental response, and multicellular growth. Zygnematophyceae and land plants share all the major enzymes for cell wall synthesis and remodifications, and gene gains shaped this toolkit. Co-expression network analyses uncover gene cohorts that unite environmental signaling with multicellular developmental programs. Our data shed light on a molecular chassis that balances environmental response and growth modulation across more than 600 million years of streptophyte evolution
A laboratory nanoseismological study on deep-focus earthquake micromechanics
No full textGlobal earthquake occurring rate displays an exponential decay down to similar to 300 km and then peaks around 550 to 600 km before terminating abruptly near 700 km. How fractures initiate, nucleate, and propagate at these depths remains one of the greatest puzzles in earth science, as increasing pressure inhibits fracture propagation. We report nanoseismological analysis on high-resolution acoustic emission (AE) records obtained during ruptures triggered by partial transformation from olivine to spinel in Mg2GeO4, an analog to the dominant mineral (Mg,Fe)(2)SiO4 olivine in the upper mantle, using state-of-the-art seismological techniques, in the laboratory. AEs' focal mechanisms, as well as their distribution in both space and time during deformation, are carefully analyzed. Microstructure analysis shows that AEs are produced by the dynamic propagation of shear bands consisting of nanograined spinel. These nanoshear bands have a near constant thickness (similar to 100 nm) but varying lengths and self-organize during deformation. This precursory seismic process leads to ultimate macroscopic failure of the samples. Several source parameters of AE events were extracted from the recorded waveforms, allowing close tracking of event initiation, clustering, and propagation throughout the deformation/transformation process. AEs follow the Gutenberg-Richter statistics with a well-defined b value of 1.5 over three orders of moment magnitudes, suggesting that laboratory failure processes are self-affine. The seismic relation between magnitude and rupture area correctly predicts AE magnitude at millimeter scales. A rupture propagation model based on strain localization theory is proposed. Future numerical analyses may help resolve scaling issues between laboratory AE events and deep-focus earthquakes
Recommended from our members
A Platform for Functional Discoveries to Optimize Algae Productivity
No full textThe development and adoption of biofuels as a renewable energy source is necessary as a replacement for petroleum-based fuel. Traditional fossil fuels are derived from a finite source and the use of which is environmentally detrimental, including contributing to increased carbon dioxide levels and global warming. Microalgae are strong candidates for the production of these biofuels as they are fast growing, tolerant of a wide variety of environmental conditions, and naturally produce valuable products.In this study I conduct an in-depth analysis of a little researched species, Nannochloris desiccata. This species was originally misidentified as belonging to the genus Chlorella, and after full genome sequencing and phylogenetic analysis was correctly classified. Growth and productivity analysis show that this is a halotolerant species showing competitive growth as compared to the highly productive species Picochlorum soloecismus and Chlorella sorokiniana. N. desiccata can produce up to 45% FAME/dry weight five-days after nitrogen depletion in marine media, showing potential as a valuable production species.To understand the mechanisms of growth I determined the diurnal cell cycle synchronization of Nannochloropsis salina. The cells in culture grow in size, including accumulation of neutral lipids and increasing chlorophyll content during the light cycle. In the last four hours of the light cycle, they replicate their DNA, either one or two times, then divide during the dark cycle. The transformation efficiency of these cells peaks as the cells are completing DNA replication and preparing to divide.Overall, the results of this study contribute to the wider field of algae biofuels by introducing the valuable properties of N. desiccata as a production species and demonstrating an approach to improving transformation efficiency
Reprogramming the Epigenome of (Pre)malignant Hematopoietic Cells with Retinoic Acid and Ascorbate
No full textEnhancing TET2 activity through genetic or pharmacologic approaches, such as ascorbate supplementation, can slow myeloid malignancy progression. However, ascorbate alone may be insufficient to fully activate TET2 in malignant cells due to pharmacokinetic constraints and the need for chromatin remodeling to enable effective epigenetic reprogramming. Here, we identify a novel mechanism to enhance TET2 activity via all-trans retinoic acid (ATRA), which induces RARA-mediated TET2 transcription in myeloid leukemia cells and synergizes with ascorbate to promote DNA hydroxymethylation and chromatin remodeling at key myeloid differentiation loci. Using Tet1/2/3-deficient mice and primary human AML models, we show that ATRA plus ascorbate more effectively induces differentiation, inhibits leukemia stem cell self-renewal in a TET2-dependent manner, and sensitizes AML cells to targeted therapies in vivo leading to improved survival. These findings support the combined use of ATRA and ascorbate as a strategy to enhance TET2 activity for the treatment of myeloid malignancies.</p
Recommended from our members
Reprogramming the Epigenome of (Pre)malignant Hematopoietic Cells with Retinoic Acid and Ascorbate
No full textEnhancing TET2 activity through genetic or pharmacologic approaches, such as ascorbate supplementation, can slow myeloid malignancy progression. However, ascorbate alone may be insufficient to fully activate TET2 in malignant cells due to pharmacokinetic constraints and the need for chromatin remodeling to enable effective epigenetic reprogramming. Here, we identify a novel mechanism to enhance TET2 activity via all-trans retinoic acid (ATRA), which induces RARA-mediated TET2 transcription in myeloid leukemia cells and synergizes with ascorbate to promote DNA hydroxymethylation and chromatin remodeling at key myeloid differentiation loci. Using Tet1/2/3-deficient mice and primary human AML models, we show that ATRA plus ascorbate more effectively induces differentiation, inhibits leukemia stem cell self-renewal in a TET2-dependent manner, and sensitizes AML cells to targeted therapies in vivo leading to improved survival. These findings support the combined use of ATRA and ascorbate as a strategy to enhance TET2 activity for the treatment of myeloid malignancies.</p
Recommended from our members
HC-HA/PTX3 from Birth Tissue Promotes Endochondral Ossification Mediated Bone Regeneration
No full textAmniotic membrane (AM) is a tissue that forms the inner portion of the placenta and is known to contain anti-inflammatory and anti-scarring properties. Due to these properties, AM has been used in many clinical applications to downregulate the inflammation, reduce scar formation, and ultimately promote wound healing. One potential clinical unmet need where AM use could be potentially beneficial is bony non-unions, wherein skeletal bone fails to heal after 9 months of injury. Based on this need, we evaluated and demonstrated AM time- and dose-dependently promotes calcification of pre-osteoblasts (MC3T3-E1 cells), a known characteristic of osteogenesis. Such effect was independent of exogenous osteogenic supplements. We further showed HC-HA/PTX3 purified from AM promoted in vitro osteogenesis of human bone marrow mesenchymal stem cells (MSCs) through an endochondral ossification mediated pathway as demonstrated by activation of BMP signaling (a known promoter of the bone morphogenic process) and promotion of chondrogenic (collagen type II and aggrecan) and osteogenic (osterix, osteocalcin, and alkaline phosphatase) markers. Cellular aggregation was demonstrated to be an early morphological event needed to promote osteogenesis and was inhibited by AMD3100 and alpha CD44 blocking antibody suggesting CXCR4 and CD44 dependence. CXCR4 and CD44 were also shown to physically interact and form a heterodimer at 5 minutes after cell seeding which further suggests their interdependence. This interaction was followed by nuclear translocation of CXCR4 at 24 hours which was inhibited by AMD3100. We further show HC-HA/PTX3 can be reconstituted in vitro and such complex has a biochemical composition similar to HC-HA/PTX3 purified from AM with presence of HC1 and high molecular weight PTX3, but not TSG-6, HC2, and HC3. We further showed cell aggregation and subsequent endochondral ossification of MSCs can be recapitulated with reconstituted HC-HA/PTX3 but not HC1, HA, or PTX3 alone
- …
