United States

In this paper, we propose Stochastic Knowledge Distillation (SKD) to obtain compact BERT-style language model dubbed SKDBERT. In each distillation iteration, SKD samples a teacher model from a pre-defined teacher team, which consists of multiple teacher models with multi-level capacities, to transfer knowledge into student model in an one-to-one manner. Sampling distribution plays an important role in SKD. We heuristically present three types of sampling distributions to assign appropriate probabilities for multi-level teacher models. SKD has two advantages: 1) it can preserve the diversities of multi-level teacher models via stochastically sampling single teacher model in each distillation iteration, and 2) it can also improve the efficacy of knowledge distillation via multi-level teacher models when large capacity gap exists between the teacher model and the student model. Experimental results on GLUE benchmark show that SKDBERT reduces the size of a BERT$_{\rm BASE}$ model by 40% while retaining 99.5% performances of language understanding and being 100% faster. 

AAAI 2023

SKDBERT: Compressing BERT via Stochastic Knowledge Distillation

learning & optimization for snlp

text classification

model compression

The purpose of the AAAI conference series is to promote research in Artificial Intelligence (AI) and foster scientific exchange between researchers, practitioners, scientists, students, and engineers across the entirety of AI and its affiliated disciplines. AAAI-23 is the Thirty-Seventh AAAI Conference on Artificial Intelligence. The theme of this conference is to create collaborative bridges within and beyond AI. Like previous AAAI conferences, AAAI-23 will feature technical paper presentations, special tracks, invited speakers, workshops, tutorials, poster sessions, senior member presentations, competitions, and exhibit programs, and two new activities: a Bridge Program and a Lab Program. Many of these activities are tailored to the theme of bridges and all are selected according to the highest standards, with additional programs for students and young researchers. 
AAAI is providing you with a conference planner, which you can use to help organize your itinerary of activities. This includes talks to attend in person, talks to attend remotely, breaks with colleagues and your site seeing activities. To access this conference planner, please go to [https://aaai-2023.takemobi.io](https://aaai-2023.takemobi.io).

In order to access this site, you need to register. If you haven't already, please register [here](https://aaai.org/Conferences/AAAI-23/registration/).


The purpose of the AAAI conference series is to promote research in Artificial Intelligence (AI) and foster scientific exchange between researchers, practitioners, scientists, students, and engineers across the entirety of AI and its affiliated disciplines.

poster

Given a set X of n points in a metric space, the problem of diversity maximization is to extract a set S of k points from X so that the diversity of S is maximized. This problem is essential in AI-related fields, such as web search, databases, recommender systems, and data mining. Although there have been extensive studies of this problem, these studies assume that X is clean. This usually does not hold, because real-world datasets usually contain outliers. The state-of-the-art algorithm for the diversity maximization problem is based on furthest point retrieval, which is too sensitive to outliers. We therefore address the problem of diversity maximization with outliers and propose two algorithms with performance guarantee. The first algorithm runs in O((k+z)n) time, guarantees 1/2-approximation, and returns no outliers, where z is the number of outliers. The second algorithm runs in O(kz) time (which is independent of n), guarantees 1/6(1+epsilon)-approximation, and returns no outliers with constant probability. We conduct experiments on real datasets to demonstrate the effectiveness and efficiency of our algorithms.

Diversity Maximization in the Presence of Outliers

Few-shot object detection, expecting detectors to detect novel classes with a few instances, has made conspicuous progress. However, the prototypes extracted by existing meta-learning based methods still suffer from insufficient representative information and lack awareness of query images, which cannot be adaptively tailored to different query images. Firstly, only the support images are involved for extracting prototypes, resulting in scarce perceptual information of query images. Secondly, all pixels of all support images are treated equally when aggregating features into prototype vectors, thus the salient objects are overwhelmed by the cluttered background. In this paper, we propose an Information-Coupled Prototype Elaboration (ICPE) method to generate specific and representative prototypes for each query image. Concretely, a conditional information coupling module is introduced to couple information from the query branch to the support branch, strengthening the query-perceptual information in support features. Besides, we design a prototype dynamic aggregation module that dynamically adjusts intra-image and inter-image aggregation weights to highlight the salient information useful for detecting query images. Experimental results on both Pascal VOC and MS COCO demonstrate that our method achieves state-of-the-art performance in almost all settings.

Breaking Immutable: Information-Coupled Prototype Elaboration for Few-Shot Object Detection

We offer a method for one-shot mask-guided image synthesis that allows controlling manipulations of a single image by inverting a quasi-robust classifier equipped with strong regularizers. Our proposed method, entitled MAGIC, leverages structured gradients from a pre-trained quasi-robust classifier to better preserve the input semantics while preserving its classification accuracy, thereby guaranteeing credibility in the synthesis.
Unlike current methods that use complex primitives to supervise the process or use attention maps as a weak supervisory signal, MAGIC aggregates gradients over the input, driven by a guide binary mask that enforces a strong, spatial prior. MAGIC implements a series of manipulations with a single framework achieving shape and location control, intense non-rigid shape deformations, and copy/move operations in the presence of repeating objects and gives users firm control over the synthesis by requiring to simply specify binary guide masks. 
Our study and findings are supported by various qualitative comparisons with the state-of-the-art on the same images sampled from ImageNet and quantitative analysis using machine perception along with a user survey of 100+ participants that endorse our synthesis quality.

MAGIC: Mask-Guided Image Synthesis by Inverting a Quasi-Robust Classifier

Generalizing a deep learning model to new domains is crucial for computer-aided medical diagnosis systems. Most existing unsupervised domain adaptation methods have made significant progress in reducing the domain distribution gap through adversarial training. However, these methods may still produce overconfident but erroneous results on unseen target images. This paper proposes a new unsupervised domain adaptation framework for cross-modality medical image segmentation. Specifically, We first introduce two data augmentation approaches to generate two sets of semantics-preserving augmented images. Based on the model's predictive consistency on these two sets of augmented images, we identify reliable and unreliable pixels. We then perform a selective entropy constraint: we minimize the entropy of reliable pixels to increase their confidence while maximizing the entropy of unreliable pixels to reduce their confidence. Based on the identified reliable and unreliable pixels, we further propose an adaptive semantic alignment module which performs class-level distribution adaptation by minimizing the distance between same class prototypes between domains, where unreliable pixels are removed to derive more accurate prototypes. We have conducted extensive experiments on the cross-modality cardiac structure segmentation task. The experimental results show that the proposed method significantly outperforms the state-of-the-art comparison algorithms. Our code will be released soon.

Unsupervised Domain Adaptation for Medical Image Segmentation by Selective Entropy Constraints and Adaptive Semantic Alignment

Most of the recent research in semi-supervised object detection follows the pseudo-labeling paradigm evolved from the semi-supervised image classification task. However, the training paradigm of the two-stage object detector inevitably makes the pseudo-label learning process for unlabeled images full of bias. Specifically, the IoU matching scheme used for selecting and labeling candidate boxes is based on the assumption that the matching source~(ground truth) is accurate enough in terms of the number of objects, object position and object category. Obviously, pseudo-labels generated for unlabeled images cannot satisfy such a strong assumption, which makes the produced training proposals extremely unreliable and thus severely spoil the follow-up training. To de-bias the training proposals generated by the pseudo-label-based IoU matching, we propose a general framework -- De-biased Teacher, which abandons both the IoU matching and pseudo labeling processes by directly generating favorable training proposals for consistency regularization between the weak/strong augmented image pairs. Moreover, a distribution-based refinement scheme is designed to eliminate the scattered class predictions of significantly low values for higher efficiency. Extensive experiments demonstrate that the proposed De-biased Teacher consistently outperforms other state-of-the-art methods on the MS-COCO and PASCAL VOC benchmarks. Source codes are available at https://github.com/wkfdb/De-biased-Teracher.

De-biased Teacher: Rethinking IoU Matching for Semi-Supervised Object Detection

Generation-based methods have captured most of the recent attention in Generalized Zero-Shot Learning (GZSL) research. However, existing literature ignores the core of this type of method and brings limited improvements in a tortuous way. In this paper, we attempt to deconstruct the generator-classifier framework and provide guidance for its improvement and extension. We first decompose the generator-learned unseen class distribution into class-level distribution and instance-level distribution. By empirically analyzing the role of these two types of distributions in solving the GZSL problem, we generalize the focus of the generation-based approach as (i) Generator learning upon attribute generalization; and (ii) learning classifier with partly biased data. We propose a simple method based on the analysis. Without complex designs, the proposed method outperforms SotAs on four public ZSL datasets, demonstrating the validity of the proposed deconstruction. The proposed method is still effective even without a generative model, which is a step toward simplifying the generator-classifier structure. Codes are available on https://github.com/cdb342/DGZ.

Deconstructed Generation-Based Zero-Shot Model

Missing values are unavoidable in many applications of machine learning and present challenges both during training and at test time. When variables are missing in recurring patterns, fitting separate pattern submodels have been proposed as a solution. However, fitting models independently does not make efficient use of all available data. Conversely, fitting a single shared model to the full data set relies on imputation which often leads to biased results when missingness depends on unobserved factors. We propose an alternative approach, called sharing pattern submodels, which i) makes predictions that are robust to missing values at test time, ii) maintains or improves the predictive power of pattern submodels, and iii) has a short description, enabling improved interpretability. Parameter sharing is enforced through sparsity-inducing regularization which we prove leads to consistent estimation. Finally, we give conditions for when a sharing model is optimal, even when both missingness and the target outcome depend on unobserved variables. Classification and regression experiments on synthetic and real-world data sets demonstrate that our models achieve a favorable tradeoff between pattern specialization and information sharing.

Sharing Pattern Submodels for Prediction with Missing Values

Current domain adaptation methods for face anti-spoofing leverage labeled source domain data and unlabeled target domain data to obtain a promising generalizable decision boundary. However, it is usually difficult for these methods to achieve a perfect domain-invariant liveness feature disentanglement, which may degrade the final classification performance by domain differences in illumination, face category, spoof type, etc. In this work, we tackle cross-scenario face anti-spoofing by proposing a novel domain adaptation method called cyclically disentangled feature translation network (CDFTN). Specifically, CDFTN generates pseudo-labeled samples that possess: 1) source domain-invariant liveness features and 2) target domain-specific content features, which are disentangled through domain adversarial training. A robust classifier is trained based on the synthetic pseudo-labeled images under the supervision of source domain labels. We further extend CDFTN for multi-target domain adaptation by leveraging data from more unlabeled target domains. Extensive experiments on several public datasets demonstrate that our proposed approach significantly outperforms the state of the art.

Cyclically Disentangled Feature Translation for Face Anti-spoofing

There has been a recent surge of interest in introducing transformers to 3D human pose estimation (HPE) due to their powerful capabilities in modeling long-term dependencies. However, existing transformer-based methods treat body joints as equally important inputs and ignore the prior knowledge of human skeleton topology in the self-attention mechanism. To tackle this issue, in this paper, we propose a Pose-Oriented Transformer (POT) with uncertainty guided refinement for 3D HPE. Specifically, we first develop novel pose-oriented self-attention mechanism and distance-related position embedding for POT to explicitly exploit the human skeleton topology. The pose-oriented self-attention mechanism explicitly models the topological interactions between body joints, whereas the distance-related position embedding encodes the distance of joints to the root joint to distinguish groups of joints with different difficulties in regression. Furthermore, we present an Uncertainty-Guided Refinement Network (UGRN) to refine pose predictions from POT, especially for the difficult joints, by considering the estimated uncertainty of each joint with uncertainty-guided sampling strategy and self-attention mechanism. Extensive experiments demonstrate that our method significantly outperforms the state-of-the-art methods with reduced model parameters on 3D HPE benchmarks such as Human3.6M and MPI-INF-3DHP. 

Pose-Oriented Transformer with Uncertainty-Guided Refinement for 2D-to-3D Human Pose Estimation

We introduce the notion of an undisputed set for abstract argumentation frameworks, which is a conflict-free set of arguments, such that its reduct contains no non-empty admissible set. We show that undisputed sets, and the stronger notion of strongly undisputed sets, provide a meaningful approach to weaken admissibility and deal with the problem of attacks from self-attacking arguments, in a similar manner as the recently introduced notion of weak admissibility. We investigate the properties of our new semantical notions and show certain relationships to classical semantics, in particular that undisputed sets are a generalisation of preferred extensions and strongly undisputed sets are a generalisation of stable extensions. We also investigate the computational complexity of standard reasoning tasks with these new notions and show that they lie on the second and third level of the polynomial hierarchy, respectively.

Premium content

SKDBERT: Compressing BERT via Stochastic Knowledge Distillation

Downloads

Next from AAAI 2023

Diversity Maximization in the Presence of Outliers

Similar lecture

Bespoke: A Block-Level Neural Network Optimization Framework for Low-Cost Deployment