Dominican Republic

NLP models are vulnerable to data poisoning attacks. One type of attack can plant a backdoor in a model by injecting poisoned examples in training, causing the victim model to misclassify test instances which include a specific pattern. Although defences exist to counter these attacks, they are specific to an attack type or pattern. In this paper, we propose a generic defence mechanism by making the training process robust to poisoning attacks through gradient shaping methods, based on differentially private training. We show that our method is highly effective in mitigating, or even eliminating, poisoning attacks on text classification, with only a small cost in predictive accuracy.

EMNLP 2021

Mitigating Data Poisoning in Text Classification with Differential Privacy

EMNLP 2021 is planned to be a hybrid event in Punta Cana, Dominican Republic, with both on-site and fully virtual participation possible. The experience for on-site participants would closely approximate a normal pre-COVID *ACL conference, with 5-6 thematically organized parallel sessions and live Q/A and interactive discussion immediately after the talks. Presentations by virtual participants will be equitably interleaved with those of on-site participants, projected on the auditorium screens as if on-site, and also followed immediately by live Q/A and interactive discussion at a time during reasonable waking hours for the virtual presenter. For all participants, on-site and virtual, who are unable to attend a session due to either time-zone issues or because they are participating in another session live, talk recordings and slides will be available online at a minimum after the live presentation (and in many cases before as well), and questions may be submitted in advance on session-specific discussion boards and answered live in session with the usual visual aids if desired.

<iframe style="width:700px;height:400px" src="https://online.fliphtml5.com/ebtyf/ceby/"  seamless="seamless" scrolling="no" frameborder="0" allowtransparency="true" allowfullscreen="true" ></iframe>

Please Note: The EMNLP registration system is not currently connected to the underline site as we are still in the process of building out EMNLP 2021. You will receive access instructions from underline the week of November 1st. 

Access is given only to EMNLP upon registration, if you have not registered please do so [here](https://2021.emnlp.org/registration).

Registered attendees will receive access the week of November 1st.

EMNLP 2021 is planned to be a hybrid event in Punta Cana, Dominican Republic, with both on-site and fully virtual participation possible.

technical paper

Linguistic representations derived from text alone have been criticized for their lack of grounding, i.e., connecting words to their meanings in the physical world. Vision-and- Language (VL) models, trained jointly on text and image or video data, have been offered as a response to such criticisms. However, while VL pretraining has shown success on multimodal tasks such as visual question answering, it is not yet known how the internal linguistic representations themselves compare to their text-only counterparts. This paper compares the semantic representations learned via VL vs. text-only pretraining for two recent VL models using a suite of analyses (clustering, probing, and performance on a commonsense question answering task) in a language-only setting. We find that the multimodal models fail to significantly outperform the text-only variants, suggesting that future work is required if multimodal pretraining is to be pursued as a means of improving NLP in general.

Does Vision-and-Language Pretraining Improve Lexical Grounding?

Existing techniques for mitigating dataset bias often leverage a biased model to identify biased instances. The role of these biased instances is then reduced during the training of the main model to enhance its robustness to out-of-distribution data. A common core assumption of these techniques is that the main model handles biased instances similarly to the biased model, in that it will resort to biases whenever available. In this paper, we show that this assumption does not hold in general. We carry out a critical investigation on two well-known datasets in the domain, MNLI and FEVER, along with two biased instance detection methods, partial-input and limited-capacity models. Our experiments show that in around a third to a half of instances, the biased model is unable to predict the main model's behavior, highlighted by the significantly different parts of the input on which they base their decisions. Based on a manual validation, we also show that this estimate is highly in line with human interpretation. Our findings suggest that down-weighting of instances detected by bias detection methods, which is a widely-practiced procedure, is an unnecessary waste of training data. We release our code to facilitate reproducibility and future research.

Don't Discard All the Biased Instances: Investigating a Core Assumption in Dataset Bias Mitigation Techniques

We present a new form of ensemble method--Devil's Advocate, which uses a deliberately dissenting model to force other submodels within the ensemble to better collaborate. Our method consists of two different training settings: one follows the conventional training process (Norm), and the other is trained by artificially generated labels (DevAdv). After training the models, Norm models are fine-tuned through an additional loss function, which uses the DevAdv model as a constraint. In making a final decision, the proposed ensemble model sums the scores of Norm models and then subtracts the score of the DevAdv model. The DevAdv model improves the overall performance of the other models within the ensemble. In addition to our ensemble framework being based on psychological background, it also shows comparable or improved performance on 5 text classification tasks when compared to conventional ensemble methods.

Devil's Advocate: Novel Boosting Ensemble Method from Psychological Findings for Text Classification

Recent evidence supports a role for coreference processing in guiding human expectations about upcoming words during reading, based on covariation between reading times and word surprisal estimated by a coreference-aware semantic processing model (Jaffe et al. 2020).The present study reproduces and elaborates on this finding by (1) enabling the parser to process subword information that might better approximate human morphological knowledge, and (2) extending evaluation of coreference effects from self-paced reading to human brain imaging data. Results show that an expectation-based processing effect of coreference is still evident even in the presence of the stronger psycholinguistic baseline provided by the subword model, and that the coreference effect is observed in both self-paced reading and fMRI data, providing evidence of the effect's robustness.

Coreference-aware Surprisal Predicts Brain Response

A popular approach to decompose the neural bases of language consists in correlating, across individuals, the brain responses to different stimuli (e.g. regular speech versus scrambled words, sentences, or paragraphs). Although successful, this 'model-free' approach necessitates the acquisition of a large and costly set of neuroimaging data. Here, we show that a model-based approach can reach equivalent results within subjects exposed to natural stimuli. We capitalize on the recently-discovered similarities between deep language models and the human brain to compute the mapping between i) the brain responses to regular speech and ii) the activations of deep language models elicited by modified stimuli (e.g. scrambled words, sentences, or paragraphs). Our model-based approach successfully replicates the seminal study of Lerner et al. (2011), which revealed the hierarchy of language areas by comparing the functional-magnetic resonance imaging (fMRI) of seven subjects listening to 7min of both regular and scrambled narratives. We further extend and precise these results to the brain signals of 305 individuals listening to 4.1 hours of narrated stories. Overall, this study paves the way for efficient and flexible analyses of the brain bases of language.

Model-based analysis of brain activity reveals the hierarchy of language in 305 subjects

Unsupervised PCFG induction models, which build syntactic structures from raw text, can be used to evaluate the extent to which syntactic knowledge can be acquired from distributional information alone. However, many state-of-the-art PCFG induction models are word-based, meaning that they cannot directly inspect functional affixes, which may provide crucial information for syntactic acquisition in child learners. This work first introduces a neural PCFG induction model that allows a clean ablation of the influence of subword information in grammar induction. Experiments on child-directed speech demonstrate first that the incorporation of subword information results in more accurate grammars with categories that word-based induction models have difficulty finding, and second that this effect is amplified in morphologically richer languages that rely on functional affixes to express grammatical relations. A subsequent evaluation on multilingual treebanks shows that the model with subword information achieves state-of-the-art results on many languages, further supporting a distributional model of syntactic acquisition.

Character-based PCFG Induction for Modeling the Syntactic Acquisition of Morphologically Rich Languages

We propose Cartography Active Learning (CAL), a novel Active Learning (AL) algorithm that exploits the behavior of the model on individual instances during training as a proxy to find the most informative instances for labeling. CAL is inspired by data maps, which were recently proposed to derive insights into dataset quality (Swayamdipta et al., 2020). We compare our method on popular text classification tasks to commonly used AL strategies, which instead rely on post-training behavior. We demonstrate that CAL is competitive to other common AL methods, showing that training dynamics derived from small seed data can be successfully used for AL. We provide insights into our new AL method by analyzing batch-level statistics utilizing the data maps. Our results further show that CAL results in a more data-efficient learning strategy, achieving comparable or better results with considerably less training data.

Cartography Active Learning

Backdoor attack introduces artificial vulnerabilities into the model by poisoning a subset of the training data via injecting triggers and modifying labels. Various trigger design strategies have been explored to attack text classifiers, however, defending such attacks remains an open problem. In this work, we propose BFClass, a novel efficient backdoor-free training framework for text classification. 
The backbone of BFClass is a pre-trained discriminator that predicts whether each token in the corrupted input was replaced by a masked language model. To identify triggers, we utilize this discriminator to locate the most suspicious token from each training sample and then distill a concise set by considering their association strengths with particular labels. To recognize the poisoned subset, we examine the training samples with these identified triggers as the most suspicious token, and check if removing the trigger will change the poisoned model's prediction. Extensive experiments demonstrate that BFClass can identify all the triggers, remove 95% poisoned training samples with very limited false alarms, and achieve almost the same performance as the models trained on the benign training data.

BFClass: A Backdoor-free Text Classification Framework

Extracting salient topics from a collection of documents can be a challenging task when
a) the amount of data is large, b) the number of topics is not known a priori, and/or c) "topic noise'' is present. We define "topic noise" as the collection of documents that are irrelevant to any coherent topic and should be filtered out. By design, most clustering algorithms (e.g. k-means, hierarchical clustering) assign all input documents to one of the available clusters, guaranteeing any topic noise to propagate into the result. To address these challenges, we present a novel algorithm, {FANATIC}, that efficiently distinguishes documents from genuine topics and those that are topic noise. We also introduce a new {R}eddit dataset to showcase {FANATIC} as it contains short, noisy data that is difficult to cluster using most clustering algorithms. We find that {FANATIC} clusters 500k {R}eddit titles (of which 20% are topic noise) in 2 minutes and achieves an {AMI} score of 0.59, in contrast with hdbscan (McInnes et al., 2017), a popular algorithm suited for this type of task, which requires over 7 hours and achieves an {AMI} of 0.03. Finally, we test {FANATIC} against a {T}witter dataset and find again that it outperforms the other algorithms with an {AMI} score of 0.60. We make our code and data publicly available.

{FANATIC}: {FA}st {N}oise-{A}ware {T}op{I}c {C}lustering

Pre-training Transformer-based models such as BERT and ELECTRA on a collection of Arabic corpora, demonstrated by both AraBERT and AraELECTRA, shows an impressive result on downstream tasks. However, pre-training Transformer-based language models is computationally expensive, especially for large-scale models. Recently, Funnel Transformer has addressed the sequential redundancy inside Transformer architecture by compressing the sequence of hidden states, leading to a significant reduction in the pre-training cost. This paper empirically studies the performance and efficiency of building an Arabic language model with Funnel Transformer and ELECTRA objective. We find that our model achieves state-of-the-art results on several Arabic downstream tasks despite using less computational resources compared to other BERT-based models.

Mitigating Data Poisoning in Text Classification with Differential Privacy

Downloads

Next from EMNLP 2021

Does Vision-and-Language Pretraining Improve Lexical Grounding?

Similar lecture

Word-Level Coreference Resolution

Mitigating Data Poisoning in Text Classification with Differential Privacy

.css-70qvj9{display:-webkit-box;display:-webkit-flex;display:-ms-flexbox;display:flex;-webkit-align-items:center;-webkit-box-align:center;-ms-flex-align:center;align-items:center;}Downloads

Next from EMNLP 2021

Does Vision-and-Language Pretraining Improve Lexical Grounding?

Similar lecture

Word-Level Coreference Resolution

Downloads