United States

Deep and large pre-trained language models (e.g., BERT, GPT-3) are state-of-the-art for various natural language processing tasks. However, the huge size of these models brings challenges to fine-tuning and online deployment due to latency and cost constraints. Existing knowledge distillation methods reduce the model size, but they may encounter difficulties transferring knowledge from the teacher model to the student model due to the limited data from the downstream tasks. In this work, we propose AD^2, a novel and effective data augmentation approach to improving the task-specific knowledge transfer when compressing large pre-trained transformer models. Different from prior methods, AD^2 performs distillation by using an enhanced training set that contains both original inputs and adversarially perturbed samples that mimic the output distribution from the teacher. 
  
Experimental results show that this method allows better transfer of knowledge from the teacher to the student during distillation, producing student models that retain 99.6\% accuracy of the teacher model while outperforming existing task-specific knowledge distillation baselines by 1.2 points on average over a variety of natural language understanding tasks. Moreover, compared with alternative data augmentation methods, such as text-editing-based approaches, AD^2 is up to 28 times faster while achieving comparable or higher accuracy. In addition, when AD^2 is combined with more advanced task-agnostic distillation, we can advance the state-of-the-art performance even more. On top of the encouraging performance, this paper also provides thorough ablation studies and analysis. The discovered interplay between KD and adversarial data augmentation for compressing pre-trained Transformers may further inspire more advanced KD algorithms for compressing even larger scale models.

AAAI 2022

Adversarial Data Augmentation for Task-Specific Knowledge Distillation of Pre-Trained Transformers

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

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

The purpose of the AAAI conference is to promote research in artificial intelligence (AI) and scientific exchange among AI researchers, practitioners, scientists, and engineers in affiliated disciplines. 

In this paper, we introduce a multi-armed bandit problem termed max-min grouped bandits, in which the arms are arranged in possibly-overlapping groups, and the goal is to find a group whose worst arm has the highest mean reward. This problem is of interest in applications such as recommendation systems, and is also closely related to widely-studied robust optimization problems. We present two algorithms based successive elimination and robust optimization, and derive upper bounds on the number of samples to guarantee finding a max-min optimal or near-optimal group, as well as an algorithm-independent lower bound. We discuss the degree of tightness of our bounds in various cases of interest, and the difficulties in deriving uniformly tight bounds.

Max-Min Grouped Bandits

Multi-view subspace clustering (MVSC) optimally integrates multiple graph structure information to improve clustering performance. Recently, many anchor-based variants are proposed to reduce the computational complexity of MVSC. Though achieving considerable acceleration, we observe that most of them adopt fixed anchor points separating from the subsequential anchor graph construction, which may adversely affect the clustering performance. In addition, post-processing is required to generate discrete clustering labels with additional time consumption. To address these issues, we propose a scalable and parameter-free MVSC method to directly output the clustering labels with optimal anchor graph, termed as Efficient One-pass Multi-view Subspace Clustering with Consensus Anchors (EOMSC-CA). Specially, we combine anchor learning and graph construction into a uniform framework to boost clustering performance. Meanwhile, by imposing a graph connectivity constraint, our algorithm directly outputs the clustering labels without any post-processing procedures as previous methods do. Our proposed EOMSC-CA is proven to be linear complexity respecting to the data size. The superiority of our EOMSC-CA over the effectiveness and efficiency is demonstrated by extensive experiments. Our code is publicly available at https://github.com/Tracesource/EOMSC-CA.

Efficient One-Pass Multi-View Subspace Clustering with Consensus Anchors

Existing Scene Text Recognition (STR) methods typically use a language model to optimize the joint probability of the 1D character sequence predicted by a visual recognition (VR) model, which ignore the 2D spatial context of visual semantics within and between character instances, making them not generalize well to arbitrary shape scene text. To address this issue, we make the first attempt to perform textual reasoning based on visual semantics in this paper. Technically, given the character segmentation maps predicted by a VR model, we construct a subgraph for each instance, where nodes represent the pixels in it and edges are added between nodes based on their spatial similarity. Then, these subgraphs are sequentially connected by their root nodes and merged into a complete graph. Based on this graph, we devise a graph convolutional network for textual reasoning (GTR) by supervising it with a cross-entropy loss. GTR can be easily plugged in representative STR models to improve their performance owing to better textual reasoning. Specifically, we construct our model, namely S-GTR, by paralleling GTR to the language model in a segmentation-based STR baseline, which can effectively exploit the visual-linguistic complementarity via mutual learning. S-GTR sets new state-of-the-art on six challenging STR benchmarks and generalizes well to multi-linguistic datasets. Code is available at https://github.com/adeline-cs/GTR.

Visual Semantics Allow for Textual Reasoning Better in Scene Text Recognition

Pose-invariant face recognition is a practically useful but challenging task. This paper introduces a novel method to learn pose-invariant feature representation without normalizing profile faces to frontal ones or learning disentangled features. We first design a novel strategy to learn pose-invariant feature embeddings by distilling the angular knowledge of frontal faces extracted by teacher network to student network, which enables the handling of faces with large pose variations. In this way, the features of faces across variant poses can cluster compactly for the same person to create a pose-invariant face representation. Secondly, we propose a Pose-Adaptive Angular Distillation loss to mitigate the negative effect of uneven distribution of face poses in the training dataset to pay more attention to the samples with large pose variations. Extensive experiments on two challenging benchmarks (IJB-A and CFP-FP) show that our approach consistently outperforms the existing methods.

Pose-Invariant Face Recognition via Adaptive Angular Distillation

Although point-based networks are demonstrated to be accurate for 3D point cloud modeling, they are still falling behind their voxel-based competitors in 3D detection. We observe that the prevailing set abstraction design for down-sampling points may maintain too much unimportant background information that can affect feature learning for detecting objects. To tackle this issue, we propose a novel set abstraction method named Semantics-Augmented Set Abstraction (SASA). Technically, we first add a binary segmentation module as the side output to help identify foreground points. Based on the estimated point-wise foreground scores, we then propose a semantics-guided point sampling algorithm to help retain more important foreground points during down-sampling. In practice, SASA shows to be effective in identifying valuable points related to foreground objects and improving feature learning for point-based 3D detection. Additionally, it is an easy-to-plug-in module and able to boost various point-based detectors, including single-stage and two-stage ones. Extensive experiments on the popular KITTI and nuScenes datasets validate the superiority of SASA, lifting point-based detection models to reach comparable performance to state-of-the-art voxel-based methods. Code is available at https://github.com/blakechen97/SASA.

SASA: Semantics-Augmented Set Abstraction for Point-Based 3D Object Detection

Most of the existing methods for search result diversification (SRD) appeal to the greedy strategy for generating diversified results, which is formulated as a sequential process of selecting documents one-by-one, and the locally optimal choice is made at each round. Unfortunately, this strategy suffers from the following shortcomings: (1) Such a one-by-one selection process is rather time-consuming for both training and inference. (2) It works well on the premise that the preceding choices are optimal or close to the optimal solution. (3) The mismatch between the objective function used in training and the final evaluation measure used in testing has not been taken into account. We propose a novel framework through direct metric optimization for SRD (referred to as MO4SRD) based on the score-and-sort strategy. Specifically, we represent the diversity score of each document that determines its rank position based on a probability distribution. These distributions over scores naturally give rise to expectations over rank positions. Armed with this advantage, we can get the differentiable variants of the widely used diversity metrics. Thanks to this, we are able to directly optimize the evaluation measure used in testing. Moreover, we have devised a novel probabilistic neural scoring function. It jointly scores candidate documents by taking into account both cross-document interaction and permutation equivariance, which makes it possible to generate a diversified ranking via a simple sorting. The experimental results on benchmark collections show that the proposed method achieves significantly improved performance over the state-of-the-art results.

Optimize What You Evaluate with: Search Result Diversification Based on Approximate Metric Optimization

A Simple Temporal Network with Uncertainty (STNU) includes real-valued variables, called time-points; binary difference constraints on those time-points; and contingent links that represent actions with uncertain durations. STNUs have been used for robot control, web-service composition, and business processes. The most important property of an STNU is called dynamic controllability (DC); and algorithms for checking this property are called DC-checking algorithms. The DC-checking algorithm for STNUs with the best worst-case time-complexity is the RULÂ¯ algorithm due to Cairo, Hunsberger and Rizzi. Its complexity is O(mn + kÂ²n + kn log n), where n is the number of time-points, m is the number of constraints, and k is the number of contingent links. It is expected that this worst-case complexity cannot be improved upon. However, this paper provides a new algorithm, called RUL2021, that improves its performance in practice by an order of magnitude, as demonstrated by a thorough empirical evaluation.

Speeding up the RUL¯ Dynamic-Controllability-Checking Algorithm for Simple Temporal Networks with Uncertainty

Detection of manipulated face images has attracted a lot of interest recently. Various schemes have been proposed to tackle this challenging problem, where the patch-based approaches are shown to be promising. However, the existing patch-based approaches tend to treat different patches equally, which do not fully exploit the patch discrepancy for effective feature learning. In this paper, we propose a Patch Diffusion (PD) module which can be integrated into the existing face manipulation detection networks to boost the performance. The PD consists of Discrepancy Patch Feature Learning (DPFL) and Attention-Aware Message Passing (AMP). The DPFL effectively learns the patch features by a newly designed Pairwise Patch Loss (PPLoss), which takes both the patch importance and correlations into consideration. The AMP diffuses the patches through attention-aware message passing in a graph network, where the attentions are explicitly computed based on the patch features learnt in DPFL. We integrate our PD module into four recent face manipulation detection networks, and carry out the experiments on four popular datasets. The results demonstrate that our PD module is able to boost the performance of the existing networks for face manipulation detection.

Patch Diffusion: A General Module for Face Manipulation Detection

Recent techniques to solve photorealistic style transfer within deep convolutional neural networks (CNNs) generally require intensive training from large-scale datasets, thus having limited applicability and poor generalization ability to unseen images or styles. To overcome this, we propose a novel framework, dubbed Deep Translation Prior (DTP), to accomplish photorealistic style transfer through test-time training on given input image pair with untrained networks, which learns an image pair-specific translation prior and thus yields better performance and generalization. Tailored for such test-time training for style transfer, we present novel network architectures, with two sub-modules of correspondence and generation modules, and loss functions consisting of contrastive content, style, and cycle consistency losses. Our framework does not require offline training phase for style transfer, which has been one of the main challenges in existing methods, but the networks are to be solely learned during test time. Experimental results prove that our framework has a better generalization ability to unseen image pairs and even outperforms the state-of-the-art methods.

Deep Translation Prior: Test-Time Training for Photorealistic Style Transfer

Pedestrian trajectory prediction is crucial in many practical applications due to the diversity of pedestrian movements, such as social interactions and individual motion behaviors. With similar observable trajectories and social environments, different pedestrians may make completely different future decisions. However, most existing methods only focus on the frequent modal of the trajectory and thus are difficult to generalize to the peculiar scenario, which leads to the decline of the multimodal fitting ability when facing similar scenarios. In this paper, we propose a complementary attention gated network (CAGN) for pedestrian trajectory prediction, in which a dual-path architecture including normal and inverse attention is proposed to capture both frequent and peculiar modals in spatial and temporal patterns, respectively. Specifically, a complementary block is proposed to guide normal and inverse attention, which are then be summed with learnable weights to get attention features by a gated network. Finally, multiple trajectory distributions are estimated based on the fused spatio-temporal attention features due to the multimodality of future trajectory. Experimental results on benchmark datasets, i.e., the ETH, and the UCY, demonstrate that our method outperforms state-of-the-art methods by 13.8% in Average Displacement Error (ADE) and 10.4% in Final Displacement Error (FDE). Code will be available at https://github.com/jinghaiD/CAGN

Downloads

Next from AAAI 2022

Max-Min Grouped Bandits

Similar lecture

Towards Zero-Shot Knowledge Distillation for Natural Language Processing

.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 AAAI 2022

Max-Min Grouped Bandits

Similar lecture

Towards Zero-Shot Knowledge Distillation for Natural Language Processing

Downloads