arXiv subjects

En-hui Yang

Publications and source records attributed to En-hui Yang.

11 recordsLinked to original sources

arXiv2026

Normalized Conditional Mutual Information Surrogate Loss for Deep Neural Classifiers

In this paper, we propose a novel information theoretic surrogate loss; normalized conditional mutual information (NCMI); as a drop in alternative to the de facto cross-entropy (CE) for training deep neural network (DNN) based classifiers. We first observe that the model's NCMI is inversely proportional to its accuracy. Building on this insight, we introduce an alternating algorithm to efficiently minimize the NCMI. Across image recognition and whole-slide imaging (WSI) subtyping benchmarks, NCMI-trained models surpass state of the art losses by substantial margins at a computational cost comparable to that of CE. Notably, on ImageNet, NCMI yields a 2.77% top-1 accuracy improvement with ResNet-50 comparing to the CE; on CAMELYON-17, replacing CE with NCMI improves the macro-F1 by 8.6% over the strongest baseline. Gains are consistent across various architectures and batch sizes, suggesting that NCMI is a practical and competitive alternative to CE.

cs.LG↗

arXiv2025

Towards Undistillable Models by Minimizing Conditional Mutual Information

A deep neural network (DNN) is said to be undistillable if, when used as a black-box input-output teacher, it cannot be distilled through knowledge distillation (KD). In this case, the distilled student (referred to as the knockoff student) does not outperform a student trained independently with label smoothing (LS student) in terms of prediction accuracy. To protect intellectual property of DNNs, it is desirable to build undistillable DNNs. To this end, it is first observed that an undistillable DNN may have the trait that each cluster of its output probability distributions in response to all sample instances with the same label should be highly concentrated to the extent that each cluster corresponding to each label should ideally collapse into one probability distribution. Based on this observation and by measuring the concentration of each cluster in terms of conditional mutual information (CMI), a new training method called CMI minimized (CMIM) method is proposed, which trains a DNN by jointly minimizing the conventional cross entropy (CE) loss and the CMI values of all temperature scaled clusters across the entire temperature spectrum. The resulting CMIM model is shown, by extensive experiments, to be undistillable by all tested KD methods existing in the literature. That is, the knockoff students distilled by these KD methods from the CMIM model underperform the respective LS students. In addition, the CMIM model is also shown to performs better than the model trained with the CE loss alone in terms of their own prediction accuracy.

cs.LG↗

arXiv2025

Coded Deep Learning: Framework and Algorithm

The success of deep learning (DL) is often achieved with large models and high complexity during both training and post-training inferences, hindering training in resource-limited settings. To alleviate these issues, this paper introduces a new framework dubbed ``coded deep learning'' (CDL), which integrates information-theoretic coding concepts into the inner workings of DL, to significantly compress model weights and activations, reduce computational complexity at both training and post-training inference stages, and enable efficient model/data parallelism. Specifically, within CDL, (i) we first propose a novel probabilistic method for quantizing both model weights and activations, and its soft differentiable variant which offers an analytic formula for gradient calculation during training; (ii) both the forward and backward passes during training are executed over quantized weights and activations, eliminating most floating-point operations and reducing training complexity; (iii) during training, both weights and activations are entropy constrained so that they are compressible in an information-theoretic sense throughout training, thus reducing communication costs in model/data parallelism; and (iv) the trained model in CDL is by default in a quantized format with compressible quantized weights, reducing post-training inference and storage complexity. Additionally, a variant of CDL, namely relaxed CDL (R-CDL), is presented to further improve the trade-off between validation accuracy and compression though requiring full precision in training with other advantageous features of CDL intact. Extensive empirical results show that CDL and R-CDL outperform the state-of-the-art algorithms in DNN compression in the literature.

cs.LG↗

arXiv2023

Deep Selector-JPEG: Adaptive JPEG Image Compression for Computer Vision in Image classification with Human Vision Criteria

With limited storage/bandwidth resources, input images to Computer Vision (CV) applications that use Deep Neural Networks (DNNs) are often encoded with JPEG that is tailored to Human Vision (HV). This paper presents Deep Selector-JPEG, an adaptive JPEG compression method that targets image classification while satisfying HV criteria. For each image, Deep Selector-JPEG selects adaptively a Quality Factor (QF) to compress the image so that a good trade-off between the Compression Ratio (CR) and DNN classifier Accuracy (Rate-Accuracy performance) can be achieved over a set of images for a variety of DNN classifiers while the MS-SSIM of such compressed image is greater than a threshold value predetermined by HV with a high probability. Deep Selector-JPEG is designed via light-weighted or heavy-weighted selector architectures. Experimental results show that in comparison with JPEG at the same CR, Deep Selector-JPEG achieves better Rate-Accuracy performance over the ImageNet validation set for all tested DNN classifiers with gains in classification accuracy between 0.2% and 1% at the same CRs while satisfying HV constraints. Deep Selector-JPEG can also roughly provide the original classification accuracy at higher CRs.

eess.IV↗

arXiv2021

High Performance Convolution Using Sparsity and Patterns for Inference in Deep Convolutional Neural Networks

Deploying deep Convolutional Neural Networks (CNNs) is impacted by their memory footprint and speed requirements, which mainly come from convolution. Widely-used convolution algorithms, im2col and MEC, produce a lowered matrix from an activation map by redundantly storing the map's elements included at horizontal and/or vertical kernel overlappings without considering the sparsity of the map. Using the sparsity of the map, this paper proposes two new convolution algorithms dubbed Compressed Pattern Overlap (CPO) and Compressed Pattern Sets (CPS) that simultaneously decrease the memory footprint and increase the inference speed while preserving the accuracy. CPO recognizes non-zero elements (NZEs) at horizontal and vertical overlappings in the activation maps. CPS further improves the memory savings of CPO by compressing the index positions of neighboring NZEs. In both algorithms, channels/regions of the activation maps with all zeros are skipped. Then, CPO/CPS performs convolution via Sparse Matrix-Vector Multiplication (SpMv) done on their sparse representations. Experimental results conducted on CPUs show that average per-layer time savings reach up to 63% and Compression Ratio (CR) up to 26x with respect to im2col. In some layers, our average per layer CPO/CPS time savings are better by 28% and CR is better by 9.2x than the parallel implementation of MEC. For a given CNN's inference, we offline select for each convolution layer the best convolutional algorithm in terms of time between either CPO or CPS and im2col. Our algorithms were selected up to 56% of the non-pointwise convolutional layers. Our offline selections yield CNN inference time savings up to 9% and CR up to 10x.

cs.CV↗

arXiv2020

Targeted Attack for Deep Hashing based Retrieval

The deep hashing based retrieval method is widely adopted in large-scale image and video retrieval. However, there is little investigation on its security. In this paper, we propose a novel method, dubbed deep hashing targeted attack (DHTA), to study the targeted attack on such retrieval. Specifically, we first formulate the targeted attack as a point-to-set optimization, which minimizes the average distance between the hash code of an adversarial example and those of a set of objects with the target label. Then we design a novel component-voting scheme to obtain an anchor code as the representative of the set of hash codes of objects with the target label, whose optimality guarantee is also theoretically derived. To balance the performance and perceptibility, we propose to minimize the Hamming distance between the hash code of the adversarial example and the anchor code under the $\ell^\infty$ restriction on the perturbation. Extensive experiments verify that DHTA is effective in attacking both deep hashing based image retrieval and video retrieval.

cs.CR↗

arXiv2013

A Universal Grammar-Based Code For Lossless Compression of Binary Trees

We consider the problem of lossless compression of binary trees, with the aim of reducing the number of code bits needed to store or transmit such trees. A lossless grammar-based code is presented which encodes each binary tree into a binary codeword in two steps. In the first step, the tree is transformed into a context-free grammar from which the tree can be reconstructed. In the second step, the context-free grammar is encoded into a binary codeword. The decoder of the grammar-based code decodes the original tree from its codeword by reversing the two encoding steps. It is shown that the resulting grammar-based binary tree compression code is a universal code on a family of probabilistic binary tree source models satisfying certain weak restrictions.

cs.IT↗

arXiv2013

New Non-asymptotic Random Channel Coding Theorems

New non-asymptotic random coding theorems (with error probability $ε$ and finite block length $n$) based on Gallager parity check ensemble and Shannon random code ensemble with a fixed codeword type are established for discrete input arbitrary output channels. The resulting non-asymptotic achievability bounds, when combined with non-asymptotic equipartition properties developed in the paper, can be easily computed. Analytically, these non-asymptotic achievability bounds are shown to be asymptotically tight up to the second order of the coding rate as $n$ goes to infinity with either constant or sub-exponentially decreasing $ε$. Numerically, they are also compared favourably, for finite $n$ and $ε$ of practical interest, with existing non-asymptotic achievability bounds in the literature in general.

cs.IT↗

arXiv2010

On Linear Operator Channels over Finite Fields

Motivated by linear network coding, communication channels perform linear operation over finite fields, namely linear operator channels (LOCs), are studied in this paper. For such a channel, its output vector is a linear transform of its input vector, and the transformation matrix is randomly and independently generated. The transformation matrix is assumed to remain constant for every T input vectors and to be unknown to both the transmitter and the receiver. There are NO constraints on the distribution of the transformation matrix and the field size. Specifically, the optimality of subspace coding over LOCs is investigated. A lower bound on the maximum achievable rate of subspace coding is obtained and it is shown to be tight for some cases. The maximum achievable rate of constant-dimensional subspace coding is characterized and the loss of rate incurred by using constant-dimensional subspace coding is insignificant. The maximum achievable rate of channel training is close to the lower bound on the maximum achievable rate of subspace coding. Two coding approaches based on channel training are proposed and their performances are evaluated. Our first approach makes use of rank-metric codes and its optimality depends on the existence of maximum rank distance codes. Our second approach applies linear coding and it can achieve the maximum achievable rate of channel training. Our code designs require only the knowledge of the expectation of the rank of the transformation matrix. The second scheme can also be realized ratelessly without a priori knowledge of the channel statistics.

cs.IT↗

arXiv2006

Constructing LDPC Codes by 2-Lifts

We propose a new low-density parity-check code construction scheme based on 2-lifts. The proposed codes have an advantage of admitting efficient hardware implementations. With the motivation of designing codes with low error floors, we present an analysis of the low-weight stopping set distributions of the proposed codes. Based on this analysis, we propose design criteria for designing codes with low error floors. Numerical results show that the resulting codes have low error probabilities over binary erasure channels.

cs.IT↗

arXiv2006

Low-Density Parity-Check Code with Fast Decoding Speed

Low-Density Parity-Check (LDPC) codes received much attention recently due to their capacity-approaching performance. The iterative message-passing algorithm is a widely adopted decoding algorithm for LDPC codes \cite{Kschischang01}. An important design issue for LDPC codes is designing codes with fast decoding speed while maintaining capacity-approaching performance. In another words, it is desirable that the code can be successfully decoded in few number of decoding iterations, at the same time, achieves a significant portion of the channel capacity. Despite of its importance, this design issue received little attention so far. In this paper, we address this design issue for the case of binary erasure channel. We prove that density-efficient capacity-approaching LDPC codes satisfy a so called "flatness condition". We show an asymptotic approximation to the number of decoding iterations. Based on these facts, we propose an approximated optimization approach to finding the codes with good decoding speed. We further show that the optimal codes in the sense of decoding speed are "right-concentrated". That is, the degrees of check nodes concentrate around the average right degree.

cs.IT↗