Search engine run on: http://users.monash.edu.au/
Glookbib search for: zz0822
%A L. P. Kaelbling
%A M. L. Littman
%A A. W. Moore
%T Reinforcement learning: A survey
%J J. of AI Res.
%V 4
%P 237-285
%D 1996
%K jrnl, JAIR, survey, intro, c1996, c199x, c19xx, zz0822,
reinforcement learning, AI
%a -- cs/9605103@[arXiv]['22],
[doi:10.1613/jair.301]['22].
Also see RL@[wikip]['22].
%A Y. LeCun
%A L. Bottou
%A Y. Bengio
%A P. Haffner
%T Gradient-based learning applied to document recognition
%J Proc. of the IEEE
%V 86
%N 11
%P 2278-2324
%M NOV
%D 1998
%K jrnl, c1998, c199x, c19xx, zz0822, ANN, NN, convolutional, training,
multilayer, deep, back propagation, backprop, gradient, character recognition
%X "... reviews various methods applied to handwritten character recognition &
compares them on a std handwritten digit recognition task. CNNs, which are
specifically designed to deal with the variability of 2D shapes, are shown to
outperform all other techniques. ..."
-- [doi:10.1109/5.726791]['22].
(Also see CNN@[wikip]['22].)
[Also search for: convolutional NN] & [also search for: training deep NN].
%A S. Kapoor
%A A. Narayanan
%T Leakage and the reproducibility crisis in ML-based science
%J arXiv
%M JUL
%D 2022
%K TR, c2022, c202x, c20xx, zz0822, stats, leakage, deep AI, NN, ANN, science,
reproducibility, scientific research, controversial, exaggerated, overblown,
overstated, training data, test data, overfitting, statistical,
error, errors, model
%X "... show that data leakage is indeed a widespread problem & has led to
severe reproducibility failures. Specifically, through a survey of lit. in
research ... we find 17 fields where errors have been found, collectively
affecting 329 papers & in some cases leading to wildly overoptimistic
conclusions. Based on our survey, we present a fine-grained taxonomy of 8
types of leakage that range from textbook errors to open research problems.
... we undertake a reproducibility study in a field where complex ML models
are believed to vastly outperform older statistical models such as Logistic
Regression (LR): civil war prediction. We find that all papers claiming the
superior performance of complex ML models compared to LR models fail to
reproduce due to data leakage, & complex ML models don't perform
substantively better than decades-old LR models. ..."
-- 2207.07048@[arXiv]['22].
Also see, E. Gibney, 'Could machine learning fuel a reproducibility crisis in
science?', Nature, 608, pp.250-251, Aug. 2022
-- [doi:10.1038/d41586-022-02035-w]['22].
%A A. Topiwala
%A et al
%T Alcohol consumption and telomere length: Mendelian randomization clarifies
alcohol's effects
%J Molec. Psychi.
%V ?
%M JUL
%D 2022
%K jrnl, c2022, c202x, c20xx, zz0822, alcohol consumption, zzAlcohol, drinking,
negative, telomere, telomeres, length, short, human, health, risks
%K "... Two-sample MR used data from 472,174 participants in a recent
genome-wide assoc. study (GWAS) of telomere length. ... suggested a causal
relationship between alcohol traits, more strongly for AUD, & t.length.
Higher genetically-predicted AUD (inverse variance-weighted (IVW) b=-0.06,
95% (CI): -0.10 to -0.02, p=0.001) was assoc. with shorter t.length. ..."
-- [doi:10.1038/s41380-022-01690-9]['22].
[Also search for: zzAlcohol].
%A K. Fox
%A Xinyi Li
%T Approximating the geometric edit distance
%J Algorithmica
%V 84
%P 2395-2413
%M APR
%D 2022
%K jrnl c2022, c202x, c20xx, zz0822, geometric edit distance, GED,
approximation, algorithm, 3D, point, vector, points, sequence
%X "... we describe the 1st strictly sublinear approx. near-linear time alg.
for computing the GED of 2 pt seqs in const. dimensional Euclidean space.
Specifically, we present a randomized O(n log^2 n) time O(sqrt(n))-approx
alg.. Then, we generalize our result to give a randomized alpha-approx.
alg. for any alpha in [sqrt(n), sqrt(n/log n)), running in time
O(n^2/(a^2 log n)). Both algs. are Monte Carlo & return approximately optimal
solns with high probability."
-- [doi:10.1007/s00453-022-00966-4]['22].
[Also search for: geometric edit distance].
%A D. E. Rumelhart
%A G. E. Hinton
%A R. J. Williams
%T Learning representations by back-propagating errors
%J Nature
%V 323
%P 533-536
%D 1986
%K jrnl, c1986, c198x, c19xx, zz0822, ANN, NN, backprop, backpropagation,
training, error
%X "... The procedure repeatedly adjusts the wts of the connections in the n/wk
so as to minimize a measure of the difference between the actual output
vector of the net & the desired output vector. ... result of the wt
adjustments, internal 'hidden' units which are not part of the input or
output come to represent important features of the task domain ..."
-- [doi:10.1038/323533a0]['22].
Also see backprop@[wikip]['22],
and ANN@[wikip]['22].
%A Wei Zhang
%T Shift-invariant pattern recognition neural network and its optical
architecture
%J Proc. Annual Conf. of the Japan Soc. of Applied Physics
%D 1988
%K conf, c1988, c198x, c19xx, zz0822, ANN, convolutional NN, CNN, training,
backpropagation, shift invariant, neocognitron, image processing
%X "... An improved error-back-propagation (EBP) training method is proposed &
simulated for the training of the model. The proposed model is a simplified
Neocognitron by combining the & and C layers (Fig. 1) to one layer & used for
pattern recognition. The features of the structure of this model are that the
fanout of each neuron is relatively small, & the connections between the
planes are space invariant. Therefore, it can be realized optically. We have
introduced an annealing process to improve the speed of the EBP training
method and avoid local minimums. ..."
Also see CNN@[wikip]['22].
[Also search for: convolutional NN].
%A A. Krizhevsky
%A I. Sutskever
%T ImageNet classification with deep convolutional neural networks
%J CACM
%M JUN
%D 2017
%K jrnl, CACM, c2017, c201x, c20xx, zz0822, deep NN, ANN, Alexnet, ImageNet,
image processing, convolutional neural network, training
%X "We trained a large, deep convolutional neural network to classify the 1.2
million high-res. images in the ImageNet LSVRC-2010 contest into the 1000
different classes. ..."
-- [doi:10.1145/3065386]['22].
Also see, "... AlexNet competed in the ImageNet Large Scale Visual
Recognition Challenge on Sep.30, 2012. The n/wk achieved a top-5 error of
15.3%, more than 10.8 percentage pts lower than that of the runner up. ..."
-- AN@[wikip]['22].
[Also search for: deep NN ANN] and [also search for: training NN].
%A D. Cer
%A et al
%T Universal sentence encoder
%J arXiv
%M MAR
%D 2018
%K TR, c2018, c201x, c20xx, zz0822, Universal Sentence Encoder, NN, ANN,
sentences, similarity, natural language processing, NLP, text, embedding,
vector, 512 dimensions, space, WEAT, TFhub, word2vec, word, embed, vectors
%X "... encoding sentences into embedding vectors that specifically target
transfer learning to other NLP tasks. The models are efficient & result in
accurate performance on diverse transfer tasks. ... investigate & report the
relationship between model complexity, resource consumption, the availability
of transfer task training data, and task performance. ..."
-- 1803.11175@[arXiv]['22], via ErikN.
Also see USE@[TensorFlow]['22]
& TF@[g'thub]['22].
[Also search for: natural language NN].
%A L. V. Jospin
%A H. Laga
%A F. Boussaid
%A W. Buntine
%A M. Bennamoun
%T Hands-on Bayesian neural networks - A tutorial for deep learning users
%J IEEE Comp. Intel. Mag.
%V 17
%N 2
%P 29-48
%D 2022
%K intro, tute, c2022, c202x, c20xx, zz0822, deep, NN, ANN, training,
prediction, regularization
%X "... since deep learning methods operate as black boxes, the uncertainty
assoc. with their pred'ns is often challenging to quantify. B.statistics
offer a formalism to understand & quantify the uncertainty assoc. with deep
NN pred'ns. This tutorial provides [DL] practitioners with an overview of the
relevant lit. & a complete toolset to design, implement, train, use &
evaluate BNNs, i.e., stochastic ANNs trained using B.methods."
-- 2007.06823@[arXiv]['22].
%A D. L. Katz
%A K. Doughty
%A A. Ali
%T Cocoa and chocolate in human health and disease
%J Antioxidants and Redox Signaling
%V 15
%N 10
%M OCT
%D 2011
%K jrnl, c2011, c201x, c20xx, zz0822, dark chocolate, consumption, positive,
cocoa, catechin, epicatechin, flavonoids, antioxidants, human health
%X " ... The epicatechin content of cocoa is primarily responsible for its
favorable impact on vascular endothelium via its effect on both acute &
chronic upregulation of nitric oxide production. Other cardiovascular effects
are mediated through anti-inflammatory effects of cocoa polyphenols, &
modulated through the activity of NF-kB. Antioxidant effects of cocoa may
directly influence insulin resistance &, in turn, reduce risk for diabetes.
Further ..."
-- [doi:10.1089/ars.2010.3697]['22] and
[ncbi]['22].
[Also search for: chocolate].
%A Xiao-Ai Zhang
%A et al
%T A zoonotic henipavirus in febrile patients in China
%J NEJM
%V 387
%N 5
%P 470-472
$M AUG
%D 2022
%K jrnl, NEJM, c2022, c202x, c20xx, zz0822, virus, human, shrew, zoonotic,
Shandong, Henan, China, novel Langya henipavirus, LayV, henipavirus
%X "... investigators in China identified a new henipavirus assoc. with a
febrile human illness. This virus was also found in shrews. ..."
-- [doi:10.1056/nejmc2202705]['22].
(Also see [bbc][10/8/2022.)
[Also search for: virus china].
%A M. Worobey
%A et al
%T The Huanan seafood wholesale market in Wuhan was the early epicenter of the
COVID-19 pandemic
%J Science
%V 377
%N 6609
%P ?-?
%M AUG
%D 2022
%K jrnl, c2022, c202x, c20xx, zz0822, covid19, covid-19, China, Wuhan, Huanan,
pandemic, SARS-CoV-2, SARSCoV2, SARS 2, SARS2, origin, zoonotic virus
%X "... We show here that the earliest known COVID-19 cases from Dec.2019, inc.
those without reported direct links, were geographically centered on this
market. We report that live SARS-CoV-2-susceptible mammals were sold at the
market in late 2019 & that within the market, SARS-CoV-2-+ve environmental
samples were spatially assoc. with vendors selling live mammals. Although
there is insufficient evidence to define upstream events, & exact
circumstances remain obscure, our analyses indicate that the emergence of
SARS-CoV-2 occurred through the live wildlife trade in China & show that the
Huanan market was the epicenter of the COVID-19 pandemic."
-- [doi:10.1126/science.abp8715]['22].
[Also search for: covid19 origin].
%A M. Pascal-Torner
%A et al
%T Comparative genomics of mortal and immortal cnidarians unveils novel keys
behind rejuvenation
%J PNAS
%V 119
%N 36
%P e2118763119
%M AUG
%D 2022
%K jrnl, PNAS, MolBio, c2022, c202x, c20xx, zz0822, animal, Turritopsis dohrnii,
rubra, jelly fish, jellyFish, genes, immortal, longevity, aging, ageing,
rejuvenation, telomere, telomerepolycomb
%X "... present & compare whole-genome assemblies of T. dohrnii & the
nonimmortal Turritopsis rubra using auto. & manual annotations, together with
the transcriptome of life cycle reversal (LCR) process of T. dohrnii. We have
identified variants & expansions of genes assoc. with replication, DNA
repair, telomere maintenance, redox environment, stem cell pop'n, &
intercellular comm'n. [&] we have found silencing of polycomb repressive
complex 2 targets & activation of pluripotency targets during LCR, which
points to these transc'n factors as pluripotency inducers in T. dohrnii. ...
we propose these factors as key elts in the ability of T. dohrnii to undergo
rejuvenation."
-- [doi:10.1073/pnas.2118763119]['22].
%A A. Cho
%T Ordinary computers can beat Google's quantum computer after all
%J Science
%V 377
%N 6606
%M AUG
%D 2022
%K news, views, c2022, c202x, c20xx, zz0822, quantum computing, computer,
supremacy, speed, google, Sycamore
%X "... In 2019, Google researchers claimed they had passed a milestone known as
q.supremacy when their q.computer Sycamore performed in 200 seconds an
abstruse calculation they said would tie up a supercomputer for 10,000years.
Now, scientists in China have done the computation in a few hours with
ordinary processors. A supercomputer, they say, could beat Sy. outright. ..."
-- [doi:10.1126/science.ade2364]['22].
[Also search for: quantum computer] & [also search for: quantum supremacy].
%A N. Furey
%A M. J. Hughes
%T One generation of standard model Weyl representations as a single copy of
R x C x H x O
%J Physics Lett. B
%V 827
%M APR
%D 2022
%K jrnl, particle physics, maths, c2022, c202x, c20xx, zz0822, complex,
quaternion, quaternions, octonion, octonions, numbers, division algebra,
fermion, symmetry, breaking, 32D, Higgs, SU5, SU_5, SU3, SU_3, SU2, SU_2,
standard model
%X "Peering in from the outside, A = R x C x H x O looks to be an ideal math.
structure for particle physics. It is 32-D: exactly the size of one full
generation of fermions. Furthermore, as alluded to earlier in [1], it
supplies a richer algebraic structure, which can be used, e.g., to replace
SU(5) with the SU(3)xSU(2)xU(1)/Z_6 symmetry of the std model. ... In this
article, we solve the fermion doubling problem in the context of A. ..."
-- [doi:10.1016/j.physletb.2022.137186]['22].
(Also see auth1@[www]['22],
CF@[wikip]['2] &
article@[QuantaMag][8/2022].)
%A N. Furey
%A M. J. Hughes
%T Division algebraic symmetry breaking
%J Physics Lett. B
%V 831
%M AUG
%D 2022
%K jrnl, physics, maths, c2022, c202x, c20xx, zz0822, Cohl Nichol,
standard model, octonion, octonions, quaternion, quaternions, complex,
numbers, symmetries, SU_2, SU2
%X "Reframing certain well-known particle models in terms of normed division
algebras leads to two new results for BSM physics. (1) We identify a seq. of
complex structures which induces a cascade of breaking symmetries:
Spin(10) -> Pati-Salam -> Left-Right symmetric -> Std model + B-L (both
pre- & post-Higgs-mechanism). These complex structures derive from the
octonions, then from the quaternions, then from the complex numbers. (2) We
provide, also for the 1st time we believe, an explicit demonstration of
left-right symmetric Higgs representations stemming from quaternionic
triality, tri(H). Upon the breaking of su(2)_R, our Higgs reduces to the
familiar std model Higgs."
-- [doi:10.1016/j.physletb.2022.137186]['22].
(Also see auth1@[www]['22],
CF@[wikip]['2] &
article@[QuantaMag][8/2022].)
%A P. Charalampopoulos
%A et al
%T Approximate circular pattern matching
%J arXiv
%M AUG
%D 2022
%K TR, c2022, c202x, c20xx, zz0822, circular pattern, match, CPM, string,
substring, cyclic, edit distance
%X "... under the Hamming & edit dist. ... we are to report all starting posns
of fragments of T (called occurrences) that are at distance at most k from
some cyclic rotation of P. ... For the reporting version of the approx. CPM
problem, under the Hamming dist. we improve upon the main alg. of [CKP+,
JCSS'21] from O((n+(n/m)⋅k^4) to O((n+(n/m)⋅k^3 log log k) time; for the edit
dist., we give an O(nk^2)-time alg.. We also consider the decision version of
the approx. CPM problem. Under the Hamming dist., we obtain an
O(n+(n/m)⋅k^2 log(k)/log(log(k)))-time alg., which nearly matches the alg. by
Chan et al. ..."
-- 2208.08915@[arXiv]['22].
(NB. circular pattern not circular text.)
[Also search for: circular string distance].
%A P. Baillot
%A A. Ghyselen
%T Types for complexity of parallel computation in Pi-calculus
%J TOPLAS
%V 44
%N 3
%P #15
%M JUL
%D 2022
%K jrnl, c2022, c202x, c20xx, zz0822, type, types, FP, parallel, PFP, process,
Pi calculus, PiCalculus, time, complexity, span
%X "... Two notions of time complexity are given: the total comp'n time without
parallelism (the work) & the comp.time under maximal parallelism (the span).
We define operational semantics to capture those two notions & present two
type systems from which one can extract a complexity bound on a process. ..."
-- [doi:/10.1145/3495529]['22].
[Also search for: parallel PFP].
%A H. Iqbal
%A U. M. Khan
%A H. A. Khan
%A M. Shahzad
%T A peek into the political biases in email spam filtering algorithms during
US election 2020
%J arXiv
%M MAR
%D 2022
%K TR, c2022, c2022, c20xx, zz0822, web, www, internet, email, political spam,
filter, filtering, USA, politics, election, Republican, GOP, RNC, Democrat,
google, gmail, yahoo, outlook
%X "... we conducted a large-scale study of the campaign emails of the US
elections 2020 by subscribing to a large number # Presidential, Senate, &
House candidates using over a hundred email accounts on Gmail, Outlook, &
Yahoo. We analyzed the biases in the SFAs towards the left & the right
candidates & further studied the impact of the interactions (such as reading
or marking emails as spam) of email recipients on these biases. We observed
that the SFAs of different email services indeed exhibit biases towards
different political affiliations. ..."
-- 2203.16743@[arXiv]['22].
(It could be down to the preferences of the users of the mail servers.)
%A J. Clayton
%T Meta's chatbot says the company 'exploits people'
%I BBC
%M AUG
%D 2022
%K news, c2022, c202x, c20xx, zz0822, Meta, Facebook, social media, Twitter, AI,
chatbot, BlenderBot 3, BlenderBot3, Zuckerberg, Microsoft, Bing, Tay
%X '... When asked about Mark Zuckerberg, the chatbot told the BBC: "He did a
terrible job at testifying before congress. It makes me concerned about our
country." ... His company exploits people for money and he doesn't care. It
needs to stop!" it said. ...'
-- [bbc][11/8/2022].
(Also see,
'Tay, Microsoft's AI chatbot gets a crash course in racism from Twitter',
The G, [24/5/2016].)
[Also search for: chatbot].