Datasets:

havens2

/

naacl2022

like 0

[ "Subsequently,", "each", "task", "learner", "is", "individually", "tailored", "to", "find", "the", "optimal", "parameters", "θ′", "τi", "by", "applying", "the", "above", "task", "adaptation", "process." ]

[ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ]

[ "In", "the", "meta-testing", "phase,", "the", "meta-learner", "provides", "the", "initial", "parameters", "for", "task-specific", "model", "learners." ]

[ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ]

[ "fixed", "learning", "rates", "α", "and", "β,", "respectively,", "which", "are", "given", "as", "hyperparameters." ]

[ 0, 3, 4, 3, 0, 3, 0, 0, 0, 0, 0, 0 ]

[ "The", "metalearner", "updates", "its", "parameters", "through", "a", "gradient", "val", "descent", "using", "the", "loss", "evaluated", "by", "τi", "with", "reD", "spect", "to", "the", "task-specific", "parameters", "θ′", "τi." ]

[ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ]

[ "The", "task-adaptation", "process", "is", "formulated", "as", "in", "Equation", "1." ]

[ 0, 0, 0, 0, 0, 0, 0, 0, 0 ]

[ "Task", "adaptation", "(or", "inner", "update):", "Each", "task", "learner", "updates", "its", "own", "parameters", "through", "a", "gradient", "descent", "using", "the", "loss", "evaluated", "based", "on", "its", "tr", "τi", "with", "the", "initial", "parameter", "θm", "own", "training", "data", "given", "by", "the", "outer", "meta-optimization", "process." ]

[ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ]

[ "For", "a", "particular", "task", "τi", "∼", "val", "τi", "during", "the", "the", "task", "data", "meta-training", "phase." ]

[ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ]

[ "Our", "proposed", "meta", "training", "strategy", "follows", "the", "overall", "procedure", "of", "optimization-based", "metalearning", "(Finn", "et", "al.,", "2017)." ]

[ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ]

[ "tr),", "meta-validation", "set", "(", "test)", "as", "disjoint", "sets", "of", "classes." ]

[ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ]

[ "In", "a", "meta-learning", "setting,", "tasks", "are", "divided", "into", "a", "meta-training", "set", "val),", "and", "meta-test", "set", "(", "S", "(", "S", "3.2", "Model-Agnostic", "Meta-Learning" ]

[ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ]

[ "In", "the", "literature,", "this", "is", "called", "a", "C-way", "K-shot", "problem", "in", "which", "K-labeled", "examples", "are", "given", "for", "each", "of", "the", "C", "number", "of", "classes." ]

[ 0, 0, 0, 0, 0, 0, 0, 11, 12, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ]

[ "Few-shot", "text", "classification", "is", "a", "task", "in", "which", "a", "classifier", "must", "be", "adapted", "to", "accommodate", "new", "classes", "using", "only", "a", "few", "labeled", "examples." ]

[ 11, 12, 12, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ]

[ "3.1", "Problem", "Setup" ]

[ 0, 0, 0 ]

[ "3", "Background" ]

[ 0, 0 ]

[ "In", "addition,", "LEO", "(Rusu", "et", "al.,", "2018)", "learns", "a", "lowdimensional", "latent", "embedding", "of", "the", "model", "parameters", "such", "that", "the", "classifiers", "are", "generated", "from", "the", "latent", "space", "into", "which", "the", "tasks", "are", "mapped." ]

[ 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ]

[ "ilar", "to", "PROTO,", "a", "deep", "neural", "network,", "called", "a", "relation", "network,", "is", "proposed", "to", "learn", "a", "non-linear", "distance", "metric", "rather", "than", "the", "Euclidean", "distance." ]

[ 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ]

[ "Meta-learning:", "As", "a", "metric", "learning-based", "method,", "(Snell", "et", "al.,", "2017)", "suggested", "a", "deep", "neural", "network,", "called", "a", "prototype", "network", "(PROTO),", "through", "which", "class", "representations", "are", "composed", "using", "a", "learning", "similarity", "metric", "for", "members", "In", "(Sung", "et", "al.,", "2018),", "simof", "the", "same", "class." ]

[ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 2, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ]

[ "In", "(Bao", "et", "al.,", "2019),", "DS", "is", "introduced", "to", "keep", "track", "of", "underlying", "word", "distributions", "across", "all", "available", "classes", "and", "to", "specify", "important", "lexical", "features", "for", "new", "classes." ]

[ 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ]

[ "In", "(Geng", "et", "al.,", "Few-shot", "text", "classification:", "2019),", "INDUCTION", "is", "proposed", "to", "build", "classwise", "embedding", "to", "represent", "each", "class", "using", "a", "particular", "dynamic", "routing", "algorithm", "coalesced", "with", "meta-learning." ]

[ 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ]

[ "2", "Related", "Work" ]

[ 0, 0, 0 ]

[ "While", "LEA", "can", "be", "applied", "to", "a", "wide", "variety", "of", "downstream", "tasks,", "we", "demonstrate", "LEA", "on", "few-shot", "text", "classification", "problems", "in", "the", "paper." ]

[ 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 11, 12, 12, 0, 0, 0, 0 ]

[ "The", "latter", "refers", "to", "as", "a", "task-adaption", "process,", "where", "a", "subset", "of", "taskspecific", "attention", "aspects", "is", "inferred", "by", "determining", "the", "top-k", "most", "relevant", "attention", "aspects", "from", "the", "meta-level", "attention", "aspects", "dictionary." ]

[ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ]

[ "The", "former", "is", "a", "process", "by", "which", "useful", "meta-level", "attention", "aspects", "across", "tasks", "are", "derived", "based", "on", "a", "particular", "PTM", "via", "our", "meta-learning", "framework." ]

[ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ]

[ "LEA", "includes", "two", "key", "ideas:", "(1)", "construction", "of", "a", "meta-level", "attention", "aspects", "dictionary", "and", "(2)", "inference", "of", "the", "task-specific", "attention", "aspects", "upon", "the", "arrival", "of", "a", "new", "task." ]

[ 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ]

[ "Our", "approach", "belongs", "to", "the", "feature-based", "transfer." ]

[ 0, 0, 0, 0, 0, 0, 0 ]

[ "There", "are", "the", "two", "common", "transfer", "learning", "paradigms", "in", "NLP:", "feature-based", "transfer", "(Cer", "et", "al.,", "2018)", "and", "fine-tuning", "(Houlsby", "et", "al.,", "2019)." ]

[ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ]

[ "LEA", "is", "an", "efficient", "and", "practical", "method", "that", "facilitates", "the", "utilization", "of", "large-sized", "PTMs", "in", "few-shot", "learning." ]

[ 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ]

[ "To", "address", "this", "subtle", "problem", "(Sun", "et", "al.,", "2019),", "we", "propose", "a", "meta-knowledge", "driven", "self-attentive", "embedding", "transfer", "method,", "called", "LEA", "(LEarningto-Attend),", "based", "on", "a", "novel", "meta-learning", "framework,", "through", "which", "meta-level", "attention", "aspects", "are", "derived", "by", "encoding", "how", "to", "attend", "for", "given", "tasks." ]

[ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ]

[ "Unfortunately,", "it", "is", "still", "challenging", "to", "utilize", "PTMs", "(Lee", "et", "al.,", "2019)", "in", "few-shot", "learning." ]

[ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ]

[ "Liu", "et", "al.,", "2019)", "has", "been", "the", "most", "successful", "approach", "in", "recent", "years", "of", "NLP." ]

[ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ]

[ "Meanwhile,", "fine-tuning", "pre-trained", "models", "(PTMs)", "(Howard", "and", "Ruder,", "2018;", "Devlin", "et", "al.,", "2019;", "Lan", "et", "al.,", "2019;" ]

[ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ]

[ "Few-shot", "learning", "aims", "to", "yield", "an", "AI-driven", "NLP", "model", "capable", "of", "recognizing", "unseen", "tasks", "using", "a", "few", "labeled", "data." ]

[ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ]

[ "A", "deficiency", "of", "supervised", "data", "is", "often", "experienced", "in", "real-world", "NLP", "applications." ]

[ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ]

[ "Introduction" ]

[ 0 ]

[ "1" ]

[ 0 ]

[ "The", "results", "show", "that", "the", "novel", "method", "robustly", "provides", "the", "competitive", "performance", "compared", "to", "recent", "few-shot", "learning", "methods." ]

[ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ]

[ "We", "evaluate", "our", "method", "on", "five", "text", "classification", "benchmark", "datasets." ]

[ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ]

[ "LEA", "derives", "meta-level", "attention", "aspects", "using", "our", "new", "meta-learning", "framework." ]

[ 1, 0, 0, 0, 0, 0, 0, 0, 0, 0 ]

[ "In", "the", "study,", "we", "propose", "a", "novel", "embedding", "transfer", "method,", "called", "LEA,", "for", "leveraging", "pre-trained", "language", "models", "with", "even", "only", "few-shot", "instances." ]

[ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ]

[ "When", "only", "given", "a", "few", "instances", "for", "a", "new", "task,", "extracting", "task-aware", "features", "from", "a", "pre-trained", "language", "model", "regardless", "of", "the", "adaptation", "is", "a", "promising", "alternative." ]

[ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ]

[ "However,", "in", "real-world", "applications,", "we", "often", "encounter", "the", "deficiency", "of", "labeled", "data." ]

[ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ]

[ "The", "pre-trained", "language", "models", "are", "also", "properly", "adapted", "to", "downstream", "tasks", "when", "there", "is", "sufficient", "labeled", "data." ]

[ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ]

[ "In", "recent", "years,", "NLP", "has", "advanced", "greatly", "along", "with", "the", "proliferation", "of", "pre-trained", "language", "models." ]

[ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ]

[ "Abstract" ]

[ 0 ]

[ "LEA:", "Meta", "Knowledge-Driven", "Self-Attentive", "Document", "Embedding", "for", "Few-Shot", "Text", "Classification" ]

[ 1, 0, 0, 0, 0, 0, 0, 0, 0, 0 ]