Publications

Self-Influence Guided Data Reweighting for Language Model Pre-training

Tolga Bolukbasi

Sriram Ganapathy

Shikhar Vashishth

Partha Talukdar

Language Models (LMs) pre-trained with self-supervision on large text corpora have become the default starting point for developing models f… (voir plus)or various NLP tasks. Once the pre-training corpus has been assembled, all data samples in the corpus are treated with equal importance during LM pre-training. However, due to varying levels of relevance and quality of data, equal importance to all the data samples may not be the optimal choice. While data reweighting has been explored in the context of task-specific supervised learning and LM fine-tuning, model-driven reweighting for pre-training data has not been explored. We fill this important gap and propose PRESENCE, a method for jointly reweighting samples by leveraging self-influence (SI) scores as an indicator of sample importance and pre-training. PRESENCE promotes novelty and stability for model pre-training. Through extensive analysis spanning multiple model sizes, datasets, and tasks, we present PRESENCE as an important first step in the research direction of sample reweighting for pre-training language models.

2023-11-30

Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing (publié)

doi.org

openreview.net

Shape-Based Measures Improve Scene Categorization

Morteza Rezanejad

John Wilder

Dirk B. Walther

Allan D. Jepson

Sven Dickinson

Kaleem Siddiqi

Converging evidence indicates that deep neural network models that are trained on large datasets are biased toward color and texture informa… (voir plus)tion. Humans, on the other hand, can easily recognize objects and scenes from images as well as from bounding contours. Mid-level vision is characterized by the recombination and organization of simple primary features into more complex ones by a set of so-called Gestalt grouping rules. While described qualitatively in the human literature, a computational implementation of these perceptual grouping rules is so far missing. In this article, we contribute a novel set of algorithms for the detection of contour-based cues in complex scenes. We use the medial axis transform (MAT) to locally score contours according to these grouping rules. We demonstrate the benefit of these cues for scene categorization in two ways: (i) Both human observers and CNN models categorize scenes most accurately when perceptual grouping information is emphasized. (ii) Weighting the contours with these measures boosts performance of a CNN model significantly compared to the use of unweighted contours. Our work suggests that, even though these measures are computed directly from contours in the image, current CNN models do not appear to extract or utilize these grouping cues.

2023-11-30

IEEE Transactions on Pattern Analysis and Machine Intelligence (publié)

doi.org

Simulation‐based effect size analysis in the absence of drug effects to inform the design of clinical trials in Alzheimer’s disease

Daniel Andrews

Douglas L Arnold

Danilo Bzdok

Simon Ducharme

Howard Chertkow

D Louis Collins

Our results suggest low false positive probabilities for the successful trials. Cohort composition, sampling, and other trial characteristic… (voir plus)s might influence treatment‐independent group difference in cognitive decline rate. With our method, future trials could target treatment effects outside the target population’s false positive range.

2023-11-30

Alzheimer's & Dementia (publié)

doi.org

Spectral Temporal Contrastive Learning

Sacha Morin

Somjit Nath

Samira Ebrahimi Kahou

Guy Wolf

Learning useful data representations without requiring labels is a cornerstone of modern deep learning. Self-supervised learning methods, pa… (voir plus)rticularly contrastive learning (CL), have proven successful by leveraging data augmentations to define positive pairs. This success has prompted a number of theoretical studies to better understand CL and investigate theoretical bounds for downstream linear probing tasks. This work is concerned with the temporal contrastive learning (TCL) setting where the sequential structure of the data is used instead to define positive pairs, which is more commonly used in RL and robotics contexts. In this paper, we adapt recent work on Spectral CL to formulate Spectral Temporal Contrastive Learning (STCL). We discuss a population loss based on a state graph derived from a time-homogeneous reversible Markov chain with uniform stationary distribution. The STCL loss enables to connect the linear probing performance to the spectral properties of the graph, and can be estimated by considering previously observed data sequences as an ensemble of MCMC chains.

2023-11-30

arXiv (prépublication)

doi.org

arxiv.org

SWEET: Weakly Supervised Person Name Extraction for Fighting Human Trafficking

Javin Liu

Hao Yu

2023-11-30

Findings of the Association for Computational Linguistics: EMNLP 2023 (publié)

doi.org

openreview.net

Syntactic Substitutability as Unsupervised Dependency Syntax

Jasper Jian

Siva Reddy

Syntax is a latent hierarchical structure which underpins the robust and compositional nature of human language. In this work, we explore th… (voir plus)e hypothesis that syntactic dependencies can be represented in language model attention distributions and propose a new method to induce these structures theory-agnostically. Instead of modeling syntactic relations as defined by annotation schemata, we model a more general property implicit in the definition of dependency relations, syntactic substitutability. This property captures the fact that words at either end of a dependency can be substituted with words from the same category. Substitutions can be used to generate a set of syntactically invariant sentences whose representations are then used for parsing. We show that increasing the number of substitutions used improves parsing accuracy on natural data. On long-distance subject-verb agreement constructions, our method achieves 79.5% recall compared to 8.9% using a previous method. Our method also provides improvements when transferred to a different parsing setup, demonstrating that it generalizes.

2023-11-30

Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing (publié)

doi.org

openreview.net

Systematic Generalization by Finetuning? Analyzing Pretrained Language Models Using Constituency Tests

Aishik Chakraborty

Jackie CK Cheung

Timothy J. O'Donnell

Constituents are groups of words that behave as a syntactic unit. Many linguistic phenomena (e.g., question formation, diathesis alternation… (voir plus)s) require the manipulation and rearrangement of constituents in a sentence. In this paper, we investigate how different finetuning setups affect the ability of pretrained sequence-to-sequence language models such as BART and T5 to replicate constituency tests — transformations that involve manipulating constituents in a sentence. We design multiple evaluation settings by varying the combinations of constituency tests and sentence types that a model is exposed to during finetuning. We show that models can replicate a linguistic transformation on a specific type of sentence that they saw during finetuning, but performance degrades substantially in other settings, showing a lack of systematic generalization. These results suggest that models often learn to manipulate sentences at a surface level unrelated to the constituent-level syntactic structure, for example by copying the first word of a sentence. These results may partially explain the brittleness of pretrained language models in downstream tasks.

2023-11-30

Proceedings of the 6th BlackboxNLP Workshop: Analyzing and Interpreting Neural Networks for NLP (publié)

doi.org

Technological Solutions to Online Toxicity: Potential and Pitfalls

Arezo Bodaghi

Benjamin C. M. Fung

Ketra A. Schmitt

Social media platforms present a perplexing duality, acting at once as sites to build community and a sense of belonging, while also giving … (voir plus)rise to misinformation, facilitating and intensifying disinformation campaigns and perpetuating existing patterns of discrimination from the physical world. The first-step platforms take in mitigating the harmful side of social media involves identifying and managing toxic content. Users produce an enormous volume of posts which must be evaluated very quickly. This is an application context that requires machine-learning (ML) tools, but as we detail in this article, ML approaches rely on human annotators, analysts, and moderators. Our review of existing methods and potential improvements indicates that neither humans nor ML can be removed from this process in the near future. However, we see room for improvement in the working conditions of these human workers.

2023-11-30

IEEE Technology and Society Magazine (publié)

doi.org

The Vault: A Comprehensive Multilingual Dataset for Advancing Code Understanding and Generation

Dung Nguyen Manh

Nam Le Hai

Anh T. V. Dau

Anh Minh Nguyen

Khanh Nghiem

Jin Guo

Nghi D. Q. Bui

We present The Vault, an open-source dataset of high quality code-text pairs in multiple programming languages for training large language m… (voir plus)odels to understand and generate code. We propose methods for thoroughly extracting samples that use both rules and deep learning to ensure that they contain high-quality pairs of code and text, resulting in a dataset of 43 million high-quality code-text pairs. We thoroughly evaluated this dataset and discovered that when used to train common code language models (such as CodeT5, CodeBERT, and CodeGen), it outperforms the same models train on other datasets such as CodeSearchNet. These evaluations included common coding tasks such as code generation, code summarization, and code search. The Vault can be used by researchers and practitioners to train a wide range of big language models that understand code. Alternatively, researchers can use our data cleaning methods and scripts to improve their own datasets. We anticipate that using The Vault to train large language models will improve their ability to understand and generate code, propelling AI research and software development forward. We are releasing our source code and a framework to make it easier for others to replicate our results.

2023-11-30

Findings of the Association for Computational Linguistics: EMNLP 2023 (publié)

doi.org

Toward Stronger Textual Attack Detectors

Pierre Colombo

Marine Picot

Nathan Noiry

Guillaume Staerman

Pablo Piantanida

The landscape of available textual adversarial attacks keeps growing, posing severe threats and raising concerns regarding the deep NLP syst… (voir plus)em's integrity. However, the crucial problem of defending against malicious attacks has only drawn the attention of the NLP community. The latter is nonetheless instrumental in developing robust and trustworthy systems. This paper makes two important contributions in this line of search: (i) we introduce LAROUSSE, a new framework to detect textual adversarial attacks and (ii) we introduce STAKEOUT, a new benchmark composed of nine popular attack methods, three datasets, and two pre-trained models. LAROUSSE is ready-to-use in production as it is unsupervised, hyperparameter-free, and non-differentiable, protecting it against gradient-based methods. Our new benchmark STAKEOUT allows for a robust evaluation framework: we conduct extensive numerical experiments which demonstrate that LAROUSSE outperforms previous methods, and which allows to identify interesting factors of detection rate variations.

2023-11-30

Findings of the Association for Computational Linguistics: EMNLP 2023 (publié)

doi.org

arxiv.org

Towards Detecting Contextual Real-Time Toxicity for In-Game Chat

Zachary Yang

Nicolas Grenon-Godbout

Reihaneh Rabbany

Real-time toxicity detection in online environments poses a significant challenge, due to the increasing prevalence of social media and gami… (voir plus)ng platforms. We introduce ToxBuster, a simple and scalable model that reliably detects toxic content in real-time for a line of chat by including chat history and metadata. ToxBuster consistently outperforms conventional toxicity models across popular multiplayer games, including Rainbow Six Siege, For Honor, and DOTA 2. We conduct an ablation study to assess the importance of each model component and explore ToxBuster's transferability across the datasets. Furthermore, we showcase ToxBuster's efficacy in post-game moderation, successfully flagging 82.1% of chat-reported players at a precision level of 90.0%. Additionally, we show how an additional 6% of unreported toxic players can be proactively moderated.

2023-11-30

Findings of the Association for Computational Linguistics: EMNLP 2023 (publié)

doi.org

openreview.net

Towards Reliable Misinformation Mitigation: Generalization, Uncertainty, and GPT-4

Kellin Pelrine

Anne Imouza

Meilina Reksoprodjo

Camille Thibault

Caleb Gupta

Joel Christoph

Reihaneh Rabbany