Erienne V. Olesh (609987), Sergiy Yakovenko (3219324), Valeriya Gritsenko (609988) · Figshare · 2015
<p>A, Error in predicting each subject's qualitative score from regressions fitted to the rest of the participants. Mean errors are expressed as % of the correct score; error bars show s.d. across 10 movements. B, Symbols show the same data as in <a href="http://www.plosone.org/article/info:doi/10.1371/journal.pone.0104487#pone-0104487-g003" target="_blank">Fig. 3B</a>; lines show regressions for datasets with one subject's data point removed. C, Histogram of intraclass correlation coefficients for relationships between individual human raters and the mean qualitative score. D, Colored lines show reducing errors as more raters score movements of the same participants per movement type, limb, and participant.</p>
Cited by 0Open access
2014
Cited by 0
Naotsugu Tsuchiya (143920), Hiroto Kawasaki (268691), Hiroyuki Oya (268694) et al. · Figshare · 2015
<p>Peak decoding performance (in A') and the latency of the decoding, defined as the first time point when the decoding became significantly above chance (FDR q>0.05).</p>
Cited by 0Open access
Keith A. Bush (5472590), Clinton D. Kilts (228932) · Figshare · 2022
Cited by 0Open access
Cécile Trottet (16557633), Thijs Vogels (16557636), Kristina Keitel (4529020) et al. · Figshare · 2023
<p>Mean of the 5 × 2 cross-validated macro F1 scores for the diagnosis prediction on the test sets. Furthermore, MoDN significantly beats at least one of the baselines for each of the individual diagnoses except for malaria.</p>
Cited by 0Open access
gamapvitor · 2016
Cited by 0
J. T. Paire, Paul Coulton, P.G. Farrell · Lecture notes in computer science · 2001
Cited by 0
Xiaoxuan Liu, Jiaxiang Yu, Park, Jongseok et al. · arXiv (Cornell University) · 2025
Speculative decoding (SD) has become a popular technique to accelerate Large Language Model (LLM) inference, yet its real-world effectiveness remains unclear as prior evaluations rely on research prototypes and unrealistically small batch sizes. We present, to our knowledge, the first systematic study of SD on a production-grade and widely deployed inference engine (vLLM), covering multiple SD variants ($n$-gram, EAGLE/EAGLE-3, Draft-Model, Multi-Token Prediction) across diverse workloads, model scales, and batch sizes. We analyze key factors governing SD performance, and quantify a theoretical upper bound on SD speedup. Our results show that verification by the target model dominates the execution, while acceptance length varies markedly across output token positions, requests, and datasets. Comparing measured performance with theoretical bounds reveals substantial gaps between observed and theoretical upper bounds, and we leverage this observation to highlight new research opportunities that our study opens up in improving SD.
Cited by 0Open access
Kai He, Gert Cauwenberghs · 2003
We propose the Binary-Analog Channel (BAC) as a model for a binary communication channel accounting for analog saturation effects, and use the model to analyze analog Viterbi decoding. We also model imperfections in the VLSI implementation of the analog decoder, to estimate practical performance. Extensive random coding simulations indicate that an analog Viterbi decoder implemented in a typical CMOS process easily outperforms its digital counterpart with 3-bit or higher soft-decision decoding resolution.
Cited by 9
Xiaoxuan Liu, Jiaxiang Yu, Park, Jongseok et al. · ArXiv.org · 2025
Speculative decoding (SD) has become a popular technique to accelerate Large Language Model (LLM) inference, yet its real-world effectiveness remains unclear as prior evaluations rely on research prototypes and unrealistically small batch sizes. We present, to our knowledge, the first systematic study of SD on a production-grade and widely deployed inference engine (vLLM), covering multiple SD variants ($n$-gram, EAGLE/EAGLE-3, Draft-Model, Multi-Token Prediction) across diverse workloads, model scales, and batch sizes. We analyze key factors governing SD performance, and quantify a theoretical upper bound on SD speedup. Our results show that verification by the target model dominates the execution, while acceptance length varies markedly across output token positions, requests, and datasets. Comparing measured performance with theoretical bounds reveals substantial gaps between observed and theoretical upper bounds, and we leverage this observation to highlight new research opportunities that our study opens up in improving SD.
Cited by 0Open access
Rachel Mediana Untung · Universitas Kristen Satya Wacana Institutional Repository (Universitas Kristen Satya Wacana) · 2013
Proceeding The 1st International Conference On Performing Arts Indonesia Institute of Arts Yogyakarta, 1-12 December 2013 Performing Arts as Cultural Identity : Problems And Challenges. p. 257-267
Cited by 0Open access
Jia Li, Gaigai Yang, Zhiqiang Zhao · 2014
In order to reduce the bit error rate of low-density parity-check codes for layered decoding algorithm and improve the decoding convergence speed, an improved layered decoding algorithm is proposed. In layered decoding algorithm, updated bit nodes information can be used by check nodes in the iteration process while in dynamic decoding algorithm check nodes update in order according to the reliability. The proposed algorithm combines advantages of the layered and dynamic decoding algorithm. Moreover, compared with the definition of remaining information for check nodes in the dynamic decoding algorithm, a new method to evaluate the reliability is defined in the proposed algorithm. Based on 1/2 code rate china mobile multimedia broadcasting standard, simulation and comparison between the conventional layered decoding algorithm and the proposed layered decoding algorithm have been carried out. The simulation results show that under the same bit error rate, the signal to noise ratio of the improved algorithm reduces nearly 0.4dB as well as convergence speed improves obviously. The proposed layered decoding algorithm reduces the bit error rate and improves the reliability of data in communication system which uses LDPC codes as the channel coding.
Cited by 1
Abdel Halim A. Zikry, Ashraf Y. Hassan, Wageda I. Shaban et al. · International Journal of Recent Technology and Engineering (IJRTE) · 2021
Low density parity checking codes (LDPC) are one of the most important issues in coding theory at present. LDPC-code are a type of linear-block LDPC-codes. Channel coding might be considered as the finest conversant and most potent components of cellular communications systems, that was employed for transmitting errors corrections imposed by noise, fading and interfering. LDPC-codes are advanced coding gain, i.e., new area in coding. the performances of LDPC-code are similar to the Shannon-limiting, this led to the usage of decoding in several applications in digital communications systems, like DVB-S2 and WLAN802.1..This paper aims to know what is LDPC,what its application and introduce encoding algorithms that gives rise to a linear encoding time and also show that the regular and irregular LDPC performance and also introduce different methods for decoding LDPC. I discuss in detail LDPC decoding algorithm: bit flipping algorithm, as a type from hard decision .belief propagation algorithm, sum product algorithm and minimum sum algorithm as examples from soft decision .I expect that at least some students or researchers involved in researching LDPC codes would find this paper helpful.
Cited by 4Open access
Babak Mahmoudi (354650), Justin C. Sanchez (137025) · Figshare · 2015
1<p>Speed is defined by the number of trials to converge to a control policy to solve the task consistently.</p>2<p>Accuracy is defined by the rate of success after convergence. For the first 100 trials of “Target 4” acquisition and “All 4 Targets” tasks that the decoder did not converge, the accuracy is computed during adaptation.</p>
Cited by 0Open access
K.J. Keuning, F.M.J. Willems, T.J. Tjalkens · TU/e Research Portal (Eindhoven University of Technology) · 1997
Cited by 0
Mahdi Bayanifar, Robert Calderbank, Olav Tirkkonen · Aaltodoc (Aalto University) · 2023
Binary Chirp (BC) codebooks consist of ${N^{\left( {{{\log }_2}N + 3} \right)/2}}$ lines in ${\mathbb{C}^N}$, equivalent up to overall phase rotations. Exploiting the underlying algebraic structure, the BCs allow suboptimal decoders with complexity N(logN)<sup>2</sup>, based on autocorrelations between the received signal and its permuted versions. We analyze the performance of these decoders in additive white Gaussian noise channels, providing lower bounds of decoding error probability, which are tight in the limits of low and high signal-to-noise ratio. Due to the autocorrelation nature of the receiver, the error probability becomes a function of order statistics of χ<sup>2</sup>-distributed random variables. Our results can be used when dimensioning communication systems where BCs are used as component codes.
Cited by 0Open access
Martin Tomlinson, Cen Jung Tjhai, Marcel Ambroze et al. · Greater South Information System · 2017
In this chapter, we discuss the performance of codes under soft and hard decision decoding. Upper and lower bounds on hard and soft decision decoding are discussed. For hard decision decoding, evaluation of the performance of specific codes shows that full decoding produces better performance than the usual bounded distance decoder. An analysis of the upper and lower union bounds on the probability of error for maximum likelihood soft decision decoding shows that contrary to hard decision decoding above relative low values of SNR per information bit, the two bounds coincide. The implications of this observation is that the soft decision decoding performance may be determined for a linear code by the number of minimum Hamming weight codewords without the need to determine the weight distribution of the code. Numerical performance comparisons are made for a wide range of different codes. It is shown that the binomial weight distribution provides a good indicative performance for codes whose weight distribution is difficult to obtain.
Cited by 0Open access
Clare Brennan, Lynn Love, Corinne Jola · 2015
Cited by 0
John H. Williamson (8917673), Melissa Quek (8917676), Iulia Popescu (8917679) et al. · Figshare · 2020
<p>The plots show the effects of varying burstiness t and average error rate f on theHorstein decoder with <i>k</i> = 8, <i>β</i> = 8 from <i>N</i> = 10000 random simulations. Increasing burstiness leads to increased uncorrected error rates but a decrease in the decisions/bit. Consequently, the backspace-corrected entry rate is largely unaffected by burstiness.</p>
Cited by 0Open access
Martin Tomlinson, Cen Jung Tjhai, Marcel Ambroze et al. · Greater South Information System · 2017
In this chapter, we discuss the performance of codes under soft and hard decision decoding. Upper and lower bounds on hard and soft decision decoding are discussed. For hard decision decoding, evaluation of the performance of specific codes shows that full decoding produces better performance than the usual bounded distance decoder. An analysis of the upper and lower union bounds on the probability of error for maximum likelihood soft decision decoding shows that contrary to hard decision decoding above relative low values of SNR per information bit, the two bounds coincide. The implications of this observation is that the soft decision decoding performance may be determined for a linear code by the number of minimum Hamming weight codewords without the need to determine the weight distribution of the code. Numerical performance comparisons are made for a wide range of different codes. It is shown that the binomial weight distribution provides a good indicative performance for codes whose weight distribution is difficult to obtain.
Cited by 0Open access