Tags

tag description color
2-dimensional classifier or regression on two variables
5gram 5gram
adam Adam optimizer
algo non-trivial algorithm implemented
alpha alpha phase orange
analysis some extra analysis done, not just giving the test results
attention attention used
backoff
bagging bagging/bootstraping used
base base model
baseline baseline solution
bernoulli Bernoulli (naive Bayes) model used
bert bert
beta beta phase green
better-than-no-model-baseline significantly better than stupid, no-model baseline (e.g. returning the majority class)
bidirectional bidirectional
bigram bigrams considered
bilstm BiLSTM model used
bow bag of words
bpe Text segmented into BPE subword units
c++ written (partially or fully) in C++
casemarker Special handling of case
challenge-preparation prepare other challenge
challenging-america Challenging America challenge
character-level Character-level
char-n-grams character n-grams
chi-square chi-square test used
cnn Convolutional Neural Network
complement Complement variant
computer-vision computer vision
crf
crm-114 CRM-114 used
data-exploration data exploration/visualization
deathmatch deathmatch
decision-tree decision tree used
diachronic Diachronic/temporal challenge
donut Donut model
dumz20-challenge UMZ 2019/2020 (stacjonarne) - konkurs
ensemble
existing some existing solution added
fairseq Fairseq used
fast-align Fast Align
faster-r-cnn Faster R-CNN
fasttext fasttext used
feature-engineering used more advanced pre-processing, feature engineering etc.
fine-tuned fine-tuned
frage FRAGE used
geval geval
glove GloVe used
goodturing
gpt2 GPT-2 used
gpt2-large GPT-2 large used
gpt2-xlarge GPT-2 xlarge used
gradient-descent gradient-descent
graph extra graph
gru Use GRU network
hashing-trick Hashing trick used
haskell written (partially or fully) in Haskell
huggingface-transformers Huggingface Transformers
hyperparam some hyperparameter modifed
improvement existing solution modified and improved as measured by the main metric
interpolation interpolation
inverted inverted
irstlm irstlm
java written (partially or fully) in Java
just-inference Just test a model without training/fine-tuning
kenlm KenLM used
k-means k-means or its variant used
kneser-ney Use Kneser-Ney
knn k nearest neighbors
knowledge-based some external source of knowledge used
language-tool LanguageTool used
large large model
left-to-right only left to right
lemmatization lemmatization used
linear-regression linear regression used
lisp written (partially or fully) in Lisp
lm a language model used
lm-loss Loss of language model used for prediction answers
locally-weighted Locally weighted variant
logistic-regression logistic regression used
lstm LSTM network
m2m-100 M2M-100 facebook model
marian Marian NMT used
mbart MBart
mbart-large-50 MBart-50 large
mert MERT (or equivalent) for Moses
mlm Masked Language Modeling
modernization diachronic modernization
moses Moses MT
ms-read-api Microsoft Read API
ms-read-api-2021-04-12 Microsoft Read API model ver. 2021-04-12
ms-read-api-2021-09-30-preview Microsoft Read API model ver. 2021-09-30-preview
multidimensional classifier or regression on many variables
multinomial multinomial (naive Bayes) model used
multiple-outs generated multiple outputs for geval
naive-bayes Naive Bayes Classifier used
neural-network neural network used
new-leader significantly better than the current top result
n-grams n-grams used
no-fine-tuning no fine-tuning
no-model-baseline significantly better than stupid, no-model baseline (e.g. returning the majority class)
non-zero non zero value for the metric
no-pretrained Trained from scratch
no-temporal No temporal information
null-model null model baseline
oddballness oddballness used
perl written (partially or fully) in Perl
plusaplha
pol Polish
postprocessing simple postprocessing
pretrained pre-trained embeddings
probabilities return probabilities not just classes
probability probability rather oddballness
python written (partially or fully) in Python 2/3
pytorch-nn Pytorch NN
r written (partially or fully) in R
randlm RandLM
random-forest Random Forest used
ready-made Machine Learning framework/library/toolkit used, algorithm was not implemented by the submitter
ready-made-model Ready-made ML model
regexp handcrafted regular expressions used
regularization some regularization used
right-to-left model working from right to left
rnn Recurrent Neural Network
roberta RoBERTa model
roberta-base RoBERTa Base
roberta-challam RoBERTa trained on Chronicling America
roberta-pl Polish Roberta
roberta-xlm Multilingual Roberta
ruby written (partially or fully) in Ruby
rule-based rule-based solution
scala written (partially or fully) in Scala
scikit-learn sci-kit learn used
self-made algorithm implemented by the submitter, no framework used
sentence-piece sentence pieces used (unigram)
simple simple solution
small small model
standards-preparations How to standards
stemming stemming used
stop-words stop words handled in a special manner
stupid simple, stupid rule-based solution
subword-regularization Use subword-regularization
supervised supervised
svm Support Vector Machines
temporal temporal information taken into account
tesseract Tesseract OCR
tetragram tetragrams
tf term frequency
tf-idf tf-idf used
timestamp timestamp considered
tokenization special tokenization used
torch (py)torch used
train train yourself an ML model
transformer Transformer model used
transformer-decoder Transformer-decoder architecture used
transformer-encoder Transformer-encoder architecture used
transformer-encoder-decoder Transformer-encoder-decoder architecture used
trigram trigrams considered
truecasing truecasing used
tuning tuning an existing model
uedin Uedin corrector
umz-2019-challenge see https://eduwiki.wmi.amu.edu.pl/pms/19umz#Dodatkowe_punkty_za_wygranie_wyzwa.2BAUQ-
unigram only unigrams considered
unks special handling of unknown words
vowpal-wabbit Vowpal Wabbit used
word2vec Word2Vec
word-level Word-level
wordnet some wordnet used
xgboost xgboost used
zumz-2019-challenge zumz competition