United States

Structure pruning is an effective method to compress and accelerate neural networks. While filter and channel pruning are preferable to other structure pruning methods in terms of realistic acceleration and hardware compatibility, pruning methods with a finer granularity, such as intra-channel pruning, are expected to be capable of yielding more compact and computationally efficient networks. Typical intra-channel pruning methods utilize a static and hand-crafted pruning granularity due to a large search space, which leaves room for improvement in their pruning performance. In this work, we introduce a novel structure pruning method, termed as dynamic structure pruning, to identify optimal pruning granularities for intra-channel pruning. In contrast to existing intra-channel pruning methods, the proposed method automatically optimizes dynamic pruning granularities in each layer while training deep neural networks. To achieve this, we propose a differentiable group learning method designed to efficiently learn a pruning granularity based on gradient-based learning of filter groups. The experimental results show that dynamic structure pruning achieves state-of-the-art pruning performance and better realistic acceleration on a GPU compared with channel pruning. In particular, it reduces the FLOPs of ResNet50 by 71.85% without accuracy degradation on the ImageNet dataset. Our code is available at https://github.com/irishev/DSP.

AAAI 2023

Dynamic Structure Pruning for Compressing CNNs

dynamic channel pruning

network pruning

model compression

ECAI2020 featured a full programme of technical papers, as well as many AI-related events of interest for researchers, students and all attendants who are interested on contemporary AI, including:




**An extensive programme of workshops, tutorials, and invited speakers**

**The Prestigious Applications of Intelligent Systems conference (PAIS)**

**The Frontiers in AI track sessions**

**The Lab To Market event**

**The Women in AI time**

**The EU Challenges forum**

… and many more!

ECAI 2020

ECAI 2020 featured a full programme of technical papers, as well as many AI-related events of interest for researchers, students and all attendants who are interested on contemporary AI.

poster

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.

Mixed Integer programs (MIPs) are typically solved by the Branch-and-Bound algorithm. Recently, Learning to imitate fast approximations of the expert strong branching heuristic has gained attention due to its success in reducing the running time for solving MIPs. However, existing learning-to-branch methods assume that the entire training data is available in a single session of training. This assumption is often not true, and if the training data is supplied in continual fashion over time, existing techniques suffer from catastrophic forgetting. In this work, we study the hitherto unexplored paradigm of Lifelong Learning to Branch on Mixed Integer Programs. To mitigate catastrophic forgetting, we propose LIMIP, which is powered by the idea of modeling an MIP instance in the form of a bipartite graph, which we map to an embedding space using a bipartite Graph Attention Network. This rich embedding space avoids catastrophic forgetting through the application of knowledge distillation and elastic weight consolidation, wherein we learn the parameters key towards retaining efficacy and are therefore protected from significant drift. We evaluate LIMIP on a series of NP-hard problems and establish that in comparison to existing baselines, LIMIP is up to 50% better when confronted with lifelong learning

LIMIP: Lifelong Learning to solve Mixed Integer Programs

In Federated Learning (FL), a common approach for aggregating local solutions across clients is periodic full model averaging. It is, however, known that different layers of neural networks can have a different degree of model discrepancy across the clients. The conventional full aggregation scheme does not consider such a difference and synchronizes the whole model parameters at once, resulting in inefficient network bandwidth consumption. Aggregating the parameters that are similar across the clients does not make meaningful training progress while increasing the communication cost. We propose FedLAMA, a layer-wise adaptive model aggregation scheme for scalable FL. FedLAMA adjusts the aggregation interval in a layer-wise manner, jointly considering the model discrepancy and the communication cost. This fine-grained aggregation strategy enables to reduce the communication cost without significantly harming the model accuracy. Our extensive empirical study shows that, as the aggregation interval increases, FedLAMA shows a remarkably smaller accuracy drop than the periodic full aggregation, while achieving comparable communication efficiency.

Layer-wise Adaptive Model Aggregation for Scalable Federated Learning

Generative Adversarial networks (GANs) have demonstrated their powerful capability of synthesizing high-resolution images, and great efforts have been made to interpret the semantics in the latent spaces of GANs. However, existing works still have the following limitations: (1) the majority of works rely on either pretrained attribute predictors or large-scale labeled datasets, which are difficult to collect in most cases, and (2) some other methods are only suitable for restricted cases, such as focusing on interpretation of human facial images using prior facial semantics. In this paper, we propose a GAN-based method called FEditNet, aiming to discover latent semantics using very few labeled data without any pretrained predictors or prior knowledge. Specifically, we reuse the knowledge from the pretrained GANs, and by doing so, avoid overfitting during the few-shot training of FEditNet. Moreover, our layer-wise objectives which take content consistency into account also ensure the disentanglement between attributes. Qualitative and quantitative results demonstrate that our method outperforms the state-of-the-art methods on various datasets including CelebA, FFHQ and LSUN.

FEditNet: Few-shot Editing of Latent Semantics in GAN Spaces

Graph meta learning aims to learn historical knowledge from training graph neural networks (GNNs) models and adapt it to down-stream learning tasks in a target graph, which has drawn increasing attention due to its ability of knowledge transfer and fast adaptation. While existing graph meta learning approaches assume the learning tasks are from the same graph domain but lack the solution for multi-domain adaptation. In this paper, we address the multi-domain generalized graph meta learning problem, which is challenging due to non Euclidean data, inequivalent feature spaces, and heterogeneous distributions. To this end, we propose a novel solution called MD-Gram for multi-domain graph generalization. It introduces an empirical graph generalization method that uses empirical vectors to form a unified expression of non-Euclidean graph data. Then it proposes a multi-domain graphs transformation approach to transform the learning tasks from multiple source-domain graphs with inequivalent feature spaces into a common domain, where graph meta learning is conducted to learn generalized knowledge. It further adopts a domain-specific GNN enhancement method to learn customized GNN model to achieve fast adaptation in the unseen target domain. Extensive experiments based on four real-world graph domain datasets show that the proposed methods significantly outperforms the state-of-the-art in multi-domain graph meta learning tasks.

Multi-Domain Generalized Graph Meta Learning

Modern object detectors are ill-equipped to incrementally learn new emerging object classes over time due to the well-known phenomenon of catastrophic forgetting. Due to data privacy or limited storage, few or no images of the old data can be stored for replay. In this paper, we design a novel One-Shot Replay (OSR) method for incremental object detection, which is an augmentation-based method. Rather than storing original images, only one object-level sample for each old class is stored to reduce memory usage significantly, and we find that copy-paste is a harmonious way to replay for incremental object detection. In the incremental learning procedure, diverse augmented samples with co-occurrence of old and new objects to existing training data are generated. To introduce more variants for objects of old classes, we propose two augmentation modules. The object augmentation module aims to enhance the ability of the detector to perceive potential unknown objects. The feature augmentation module explores the relations between new and old classes and augments the feature space via analogy. Extensive experimental results on VOC2007 and COCO demonstrate that OSR can outperform the state-of-the-art incremental object detection methods without using extra wild data.

One-Shot Replay: Boosting Incremental Object Detection via Retrospecting One Object

There are many applications that benefit from computing the exact divergence between 2 discrete probability measures, including machine learning. Unfortunately, in the absence of any assumptions on the structure or independencies within these distributions, computing the divergence between them is an intractable problem in high dimensions. We show that we are able to compute a wide family of functionals and divergences, such as the alpha-beta divergence, between two decomposable models, i.e~chordal Markov networks, in time exponential to the treewidth of these models. The alpha-beta divergence is a family of divergences that include popular divergences such as the Kullback-Leibler divergence, the Hellinger distance, and the chi-squared divergence. Thus, we can accurately compute the exact values of any of this broad class of divergences to the extent to which we can accurately model the two distributions using decomposable models.


Computing Divergences between Discrete Decomposable Models

This paper investigates a new yet challenging problem called Reverse $k$-Maximum Inner Product Search (R$k$MIPS). Given a query (item) vector, a set of item vectors, and a set of user vectors, the problem of R$k$MIPS aims to find a set of user vectors whose inner products with the query vector are one of the $k$ largest among the query and item vectors. We propose the first subquadratic-time algorithm, i.e., Shifting-aware Asymmetric Hashing (SAH), to tackle the R$k$MIPS problem. To speed up the Maximum Inner Product Search (MIPS) on item vectors, we design a shifting-invariant asymmetric transformation and develop a novel sublinear-time Shifting-Aware Asymmetric Locality Sensitive Hashing (SA-ALSH) scheme. Furthermore, we devise a new blocking strategy based on the Cone-Tree to effectively prune user vectors (in a batch). We prove that SAH achieves a theoretical guarantee for solving the RMIPS problem. Experimental results on five real-world datasets show that SAH runs 4$\sim$8$\times$ faster than the state-of-the-art methods for R$k$MIPS while achieving F1-scores of over 90\%. The code is available at \url{https://github.com/HuangQiang/SAH}.

SAH: Shifting-aware Asymmetric Hashing for Reverse k-Maximum Inner Product Search

Error correction in automatic speech recognition (ASR) aims to correct those incorrect words in sentences generated by ASR models. Since recent ASR models usually have low word error rate (WER), to avoid affecting originally correct tokens, error correction models should only modify incorrect words, and therefore detecting incorrect words is important for error correction. Previous works on error correction either implicitly detect error words through target-source attention or CTC (connectionist temporal classification) loss, or explicitly locate specific deletion/substitution/insertion errors. However, implicit error detection does not provide clear signal about which tokens are incorrect and explicit error detection suffers from low detection accuracy. In this paper, we propose SoftCorrect with a soft error detection mechanism to avoid the limitations of both explicit and implicit error detection. Specifically, we first detect whether a token is correct or not through a probability produced by a dedicatedly designed language model, and then design a constrained CTC loss that only duplicates the detected incorrect tokens to let the decoder focus on the correction of error tokens. Compared with implicit error detection with CTC loss, SoftCorrect provides explicit signal about which words are incorrect and thus does not need to duplicate every token but only incorrect tokens; compared with explicit error detection, SoftCorrect does not detect specific deletion/substitution/insertion errors but just leaves it to CTC loss. Experiments on AISHELL-1 and Aidatatang datasets show that SoftCorrect achieves 26.1% and 9.4% CER reduction respectively, outperforming previous works by a large margin, while still enjoying fast speed of parallel generation.

SoftCorrect: Error Correction with Soft Detection for Automatic Speech Recognition

Robust training of neural networks has so far been developed in the context of white noise perturbations to limit their susceptibility to adversarial attacks. However, in applications neural networks need to be robust to a wider range of input perturbations, including contrast, brightness, and beyond. We here introduce the problem of training neural networks such that they are robust against a class of smooth intensity perturbations modelled by bias fields. We first develop an approach towards this goal based on a State-of-the-Art (SoA) robust training method utilising Interval Bound Propagation (IBP). We analyse the resulting algorithm and observe that IBP often produces very loose bounds for bias field perturbations, which may be detrimental to training. We propose an alternative approach based on Symbolic Interval Propagation (SIP), which usually results in significantly tighter bounds than IBP. We present RobNet, a tool implementing these approaches for bias field robust training. In experiments networks trained with the SIP-based approach achieved up to 31% higher certified robustness while also maintaining a better accuracy than networks trained with the IBP approach. 

Premium content

Dynamic Structure Pruning for Compressing CNNs

Downloads

Next from AAAI 2023

LIMIP: Lifelong Learning to solve Mixed Integer Programs

Similar lecture

RPN: A Residual Pooling Network for Efficient Federated Learning

Stay up to date with the latest Underline news!

PRESENTATIONS

CONFERENCES

COMPANY

RESOURCES