In this paper we describe a method to perform sequencediscriminative training of neural network acoustic models without the need for frame-level cross-entropy pre-training. We use the lattice-free version of the maximum mutual information (MMI) criterion: LF-MMI. To make its computation feasible we use a phone n-gram language model, in place of the word language model. To further reduce its space and time complexity we compute the objective function using neural network outputs at one third the standard frame rate. These changes enable us to perform the computation for the forward-backward algorithm on GPUs. Further the reduced output frame-rate also provides a significant speed-up during decoding. We present results on 5 different LVCSR tasks with training data ranging from 100 to 2100 hours. Models trained with LF-MMI provide a relative word error rate reduction of ∼11.5%, over those trained with cross-entropy objective function, and ∼8%, over those trained with cross-entropy and sMBR objective functions. A further reduction of ∼2.5%, relative, can be obtained by fine tuning these models with the word-lattice based sMBR objective function.
Protein misfolding and aggregation is observed in many amyloidogenic diseases affecting either the central nervous system or a variety of peripheral tissues. Structural and dynamic characterization of all species along the pathways from monomers to fibrils is challenging by experimental and computational means because they involve intrinsically disordered proteins in most diseases. Yet understanding how amyloid species become toxic is the challenge in developing a treatment for these diseases. Here we review what computer, in vitro, in vivo and pharmacological experiments tell us about the accumulation and deposition of the oligomers of the (Aβ, tau), α-synuclein, IAPP and superoxide dismutase 1 proteins, which have been the mainstream concept underlying Alzheimer's disease (AD), Parkinson's disease (PD), type II diabetes (T2D) and amyotrophic lateral sclerosis (ALS) research, respectively for over many years.While SOD1 is a globular protein with a well-defined 3D structure, the Aβ, tau and α-synuclein proteins belong to the class of intrinsically disordered proteins (IDPs). IDPs are also known to play a critical role in many cellular functions such as signal transduction, cell growth, binding with DNA and RNA, and transcription, and are implicated in the development of cardiovascular problems and cancers 29 . The IDPs involved in neurodegenerative diseases have a few aggregation-prone regions and overall all IDPs have a low mean hydrophobicity and a high mean net charge 30 .IDPs are structurally flexible and lack stable secondary structures in aqueous solution. When isolated, they behave as polymers in a good solvent and their radii of gyration are well described by the Flory scaling law. 31 The insolubility and high self-assembly propensity of IDPs implicated in degenerative diseases have prevented high-resolution structural determination by solution nuclear magnetic resolution (NMR) and X-ray diffraction experiments. Local information at all aggregation steps can be, however, obtained by chemical shifts, residual coupling constants, and J-couplings from NMR, exchange hydrogen/deuterium (H/D) NMR, Raman spectroscopy; and secondary structure from fast Fourier infrared spectroscopy (FTIR) or circular dichroism (CD). Long-range tertiary contacts can be deduced from paramagnetic relaxation enhancement (PRE) NMR spectroscopy and single molecule Förster resonance energy transfer (sm-FRET), and short-range distance contacts can be extracted by cross linked residues determined by mass spectrometry (MS). Low-resolution 3D information of monomers and oligomers can be obtained by ion-mobility mass-spectrometry data (IM/MS) providing cross-collision sections, dynamic light scattering (DLS), pulse field gradient NMR spectroscopy and fluorescence correlation spectroscopy (FCS) providing hydrodynamics radius, small-angle X-ray scattering (SAXS) and small-angle neutron scattering (SANS), atomic force microscopy (AFM) and transmission electron microscopy (TEM) providing height features of the aggregates, as reported by some o...
We determine the phylogenetic backbone of the East Asian mtDNA tree by using published complete mtDNA sequences and assessing both coding and control region variation in 69 Han individuals from southern China. This approach assists in the interpretation of published mtDNA data on East Asians based on either control region sequencing or restriction fragment length polymorphism (RFLP) typing. Our results confirm that the East Asian mtDNA pool is locally region-specific and completely covered by the two superhaplogroups M and N. The phylogenetic partitioning based on complete mtDNA sequences corroborates existing RFLP-based classification of Asian mtDNA types and supports the distinction between northern and southern populations. We describe new haplogroups M7, M8, M9, N9, and R9 and demonstrate by way of example that hierarchically subdividing the major branches of the mtDNA tree aids in recognizing the settlement processes of any particular region in appropriate time scale. This is illustrated by the characteristically southern distribution of haplogroup M7 in East Asia, whereas its daughter-groups, M7a and M7b2, specific for Japanese and Korean populations, testify to a presumably (pre-)Jomon contribution to the modern mtDNA pool of Japan.
The last decade has witnessed dramatic growth in the number of reactions catalyzed by electrophilic gold complexes. While proposed mechanisms often invoke the intermediacy of gold-stabilized cationic species, the nature of bonding in these intermediates remains unclear. Herein, we propose that the carbon-gold bond in these intermediates is comprised of varying degrees of both σ and π-bonding; however, the overall bond order is generally less than or equal to unity. The bonding in a given gold-stabilized intermediate, and the position of this intermediate on a continuum ranging from gold-stabilized singlet carbene to gold-coordinated carbocation, is dictated by the carbene substituents and the ancillary ligand. Experiments show that the correlation between bonding and reactivity is reflected in the yield of gold-catalyzed cyclopropanation reactions.
Time Delay Neural Networks (TDNNs), also known as onedimensional Convolutional Neural Networks (1-d CNNs), are an efficient and well-performing neural network architecture for speech recognition. We introduce a factored form of TDNNs (TDNN-F) which is structurally the same as a TDNN whose layers have been compressed via SVD, but is trained from a random start with one of the two factors of each matrix constrained to be semi-orthogonal. This gives substantial improvements over TDNNs and performs about as well as TDNN-LSTM hybrids.
The voltammetric separation of dopamine and ascorbic acid was studied with cyclic voltammetry at two kinds of carbon nanotube-modified electrodes (coated and intercalated). The anodic peak difference reached 270 mV under the present conditions. The separation mechanism and effect factors were carefully studied. Using various types of surfactants as coating dispersants of carbon nanotubes, it was demonstrated that the charge nature of the surfactants had a strong effect on the electrochemical behavior of dopamine and ascorbic acid. When the oxidation solution of carbon nanotubes was changed from the most commonly used mixed concentrated nitric acid and sulfuric acid (1 + 3 v/v) to dilute nitric acid and to hydrochloric acid, the anodic peak separation value of dopamine and ascorbic acid increased significantly, and it was shown that carboxylic acid groups attached to the carbon nanotubes were an adverse factor for the discrimination of DA from AA. These results indicated that the resolution of DA and AA was mainly attributable to the stereo porous interfacial layer formed from aggregated pores and inner cavities of the carbon nanotubes. The modified electrodes exhibited an attractive ability to measure DA and AA simultaneously and showed good stability and reproducibility.
Partial absence of the sacrum is a rare congenital defect which also occurs as an autosomal dominant trait; association with anterior meningocoele, presacral teratoma and anorectal abnormalities constitutes the Currarino triad (MIM 176450). Malformation at the caudal end of the developing notochord at approximately Carnegie stage 7 (16 post-ovulatory days), which results in aberrant secondary neurulation, can explain the observed pattern of anomalies. We previously reported linkage to 7q36 markers in two dominantly inherited sacral agenesis families. We now present data refining the initial subchromosomal localization in several additional hereditary sacral agenesis (HSA) families. We excluded several candidate genes before identifying patient-specific mutations in a homeobox gene, HLXB9, which was previously reported to map to 1q41-q42.1 and to be expressed in lymphoid and pancreatic tissues.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.
hi@scite.ai
334 Leonard St
Brooklyn, NY 11211
Copyright © 2024 scite LLC. All rights reserved.
Made with 💙 for researchers
Part of the Research Solutions Family.