Efficient Estimation of Influence of a Training Instance

Understanding the influence of a training instance on a neural network model leads to improving interpretability. However, it is difficult and inefficient to evaluate the influence, which shows how a model’s prediction would be changed if a training instance were not used. In this paper, we propose an efficient method for estimating the influence. Our method is inspired by dropout, which zero-masks a sub-network and prevents the sub-network from learning each training instance. By switching between dropout masks, we can use sub-networks that learned or did not learn each training instance and estimate its influence. Through experiments with BERT and VGGNet on classification datasets, we demonstrate that the proposed method can capture training influences, enhance the interpretability of error predictions, and cleanse the training dataset for improving generalization.

[1]  Christopher Potts,et al.  Recursive Deep Models for Semantic Compositionality Over a Sentiment Treebank , 2013, EMNLP.

[2]  Takanori Maehara,et al.  Data Cleansing for Models Trained with SGD , 2019, NeurIPS.

[3]  Ali Farhadi,et al.  Supermasks in Superposition , 2020, NeurIPS.

[4]  Sosuke Kobayashi,et al.  All Word Embeddings from One Embedding , 2020, NeurIPS.

[5]  Percy Liang,et al.  Understanding Black-box Predictions via Influence Functions , 2017, ICML.

[6]  Sameer Singh,et al.  “Why Should I Trust You?”: Explaining the Predictions of Any Classifier , 2016, NAACL.

[7]  Nitish Srivastava,et al.  Improving neural networks by preventing co-adaptation of feature detectors , 2012, ArXiv.

[8]  Richard Socher,et al.  Pointer Sentinel Mixture Models , 2016, ICLR.

[9]  Mike Schuster,et al.  Japanese and Korean voice search , 2012, 2012 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP).

[10]  Mona Attariyan,et al.  Parameter-Efficient Transfer Learning for NLP , 2019, ICML.

[11]  R. Cook.  Detection of influential observation in linear regression , 2000 .

[12]  R'emi Louf,et al.  HuggingFace's Transformers: State-of-the-art Natural Language Processing , 2019, ArXiv.

[13]  Ming-Wei Chang,et al.  BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding , 2019, NAACL.

[14]  George Kurian,et al.  Google's Neural Machine Translation System: Bridging the Gap between Human and Machine Translation , 2016, ArXiv.

[15]  Samuel R. Bowman,et al.  A Broad-Coverage Challenge Corpus for Sentence Understanding through Inference , 2017, NAACL.

[16]  Andrew Zisserman,et al.  Very Deep Convolutional Networks for Large-Scale Image Recognition , 2014, ICLR.

[17]  Yoshua Bengio,et al.  A Neural Probabilistic Language Model , 2003, J. Mach. Learn. Res..

[18]  Christopher D. Manning,et al.  Stanza: A Python Natural Language Processing Toolkit for Many Human Languages , 2020, ACL.

[19]  Xiang Zhang,et al.  Character-level Convolutional Networks for Text Classification , 2015, NIPS.

[20]  Yulia Tsvetkov,et al.  Explaining Black Box Predictions and Unveiling Data Artifacts through Influence Functions , 2020, ACL.

[21]  Rico Sennrich,et al.  Neural Machine Translation of Rare Words with Subword Units , 2015, ACL.

[22]  John Blitzer,et al.  Biographies, Bollywood, Boom-boxes and Blenders: Domain Adaptation for Sentiment Classification , 2007, ACL.

[23]  Lorenzo Porzi,et al.  Dropout distillation , 2016, ICML.

[24]  Stephen J. Wright,et al.  Training Set Debugging Using Trusted Items , 2018, AAAI.

[25]  Nitish Srivastava,et al.  Dropout: a simple way to prevent neural networks from overfitting , 2014, J. Mach. Learn. Res..

[26]  Alex Krizhevsky,et al.  Learning Multiple Layers of Features from Tiny Images , 2009 .

[27]  Philip Bachman,et al.  Learning with Pseudo-Ensembles , 2014, NIPS.

[28]  Pierre Baldi,et al.  The dropout learning algorithm , 2014, Artif. Intell..