arXiv subjects

Jun'ichi Takeuchi

Publications and source records attributed to Jun'ichi Takeuchi.

11 recordsLinked to original sources

arXiv2026

Improving the Robustness/Accuracy Tradeoff Against Adversarial Attacks Using Information Bottleneck Distillation Through Dual Teachers

Deep neural networks (DNNs) have achieved remarkable success in classical machine learning problems. However, they are known to be vulnerable to adversarial attacks. Countermeasures proposed in the literature, notably Information Bottleneck Distillation (IBD) introduced by Kuang et al., degrade the classification accuracy on clean inputs while improving the robustness to adversarial inputs. In this work, we extend the IBD framework by introducing an extra teacher model (clean teacher) trained with only clean inputs, into the distillation process from a robust teacher model trained by adversarial training. The features of both clean and robust teachers are transferred to the student through a cross-layer attention matrix. Experimental results on the CIFAR-10 and CIFAR-100 datasets show that the proposed method improves classification accuracy on clean samples compared to the original IBD, while maintaining similar accuracy on adversarial samples. Furthermore, our methods are competitive with state-of-the-art approaches, including the recent dual-teacher distillation framework B-MTARD, particularly in terms of the harmonic mean between clean and robust accuracy. We also analyze the impact of different training settings that have different influences on the attention module.

cs.LG↗

arXiv2025

Dynamical Properties of Dense Associative Memory

Dense associative memory, a fundamental instance of modern Hopfield networks, can store a large number of memory patterns as equilibrium states of recurrent networks. While the stationary-state storage capacity has been investigated, its dynamical properties have not yet been discussed. In this paper, we analyze the dynamics using an exact approach based on generating functional analysis. We show results on convergence properties of memory retrieval, such as the convergence time and the size of the attraction basins. Our analysis enables a quantitative evaluation of the convergence time and the storage capacity of dense associative memory, which is useful for model design. Unlike the traditional Hopfield model, the retrieval of a pattern does not act as additional noise to itself, suggesting that the structure of modern networks makes recall more robust. Furthermore, the methodology addressed here can be applied to other energy-based models, and thus has the potential to contribute to the design of future architectures.

cond-mat.dis-nn↗

arXiv2025

Neural Tangent Kernels and Fisher Information Matrices for Simple ReLU Networks with Random Hidden Weights

Fisher information matrices and neural tangent kernels (NTK) for 2-layer ReLU networks with random hidden weight are argued. We discuss the relation between both notions as a linear transformation and show that spectral decomposition of NTK with concrete forms of eigenfunctions with major eigenvalues. We also obtain an approximation formula of the functions presented by the 2-layer neural networks.

cs.LG↗

arXiv2025

Asymptotically Minimax Regret by Bayes Mixtures

We study the problems of data compression, gambling and prediction of a sequence $x^n=x_1x_2...x_n$ from an alphabet ${\cal X}$, in terms of regret and expected regret (redundancy) with respect to various smooth families of probability distributions. We evaluate the regret of Bayes mixture distributions compared to maximum likelihood, under the condition that the maximum likelihood estimate is in the interior of the parameter space. For general exponential families (including the non-i.i.d.\ case) the asymptotically mimimax value is achieved when variants of the prior of Jeffreys are used. %under the condition that the maximum likelihood estimate is in the interior of the parameter space. Interestingly, we also obtain a modification of Jeffreys prior which has measure outside the given family of densities, to achieve minimax regret with respect to non-exponential type families. This modification enlarges the family using local exponential tilting (a fiber bundle). Our conditions are confirmed for certain non-exponential families, including curved families and mixture families (where either the mixture components or their weights of combination are parameterized) as well as contamination models. Furthermore for mixture families we show how to deal with the full simplex of parameters. These results also provide characterization of Rissanen's stochastic complexity.

cs.IT↗

arXiv2025

Sparse Superposition Codes with Binomial Dictionary are Capacity-Achieving with Maximum Likelihood Decoding

It is known that sparse superposition codes asymptotically achieve the channel capacity over the additive white Gaussian noise channel with both maximum likelihood decoding and efficient decoding (Joseph and Barron in 2012, 2014). Takeishi et al. (in 2014, 2019) demonstrated that these codes can also asymptotically achieve the channel capacity with maximum likelihood decoding when the dictionary is drawn from a Bernoulli distribution. In this paper, we extend these results by showing that the dictionary distribution can be naturally generalized to the binomial distribution.

cs.IT↗

arXiv2024

Risk Bounds on MDL Estimators for Linear Regression Models with Application to Simple ReLU Neural Networks

To investigate the theoretical foundations of deep learning from the viewpoint of the minimum description length (MDL) principle, we analyse risk bounds of MDL estimators based on two-stage codes for simple two-layers neural networks (NNs) with ReLU activation. For that purpose, we propose a method to design two-stage codes for linear regression models and establish an upper bound on the risk of the corresponding MDL estimators based on the theory of MDL estimators originated by Barron and Cover (1991). Then, we apply this result to the simple two-layers NNs with ReLU activation which consist of $d$ nodes in the input layer, $m$ nodes in the hidden layer and one output node. Since the object of estimation is only the $m$ weights from the hidden layer to the output node in our setting, this is an example of linear regression models. As a result, we show that the redundancy of the obtained two-stage codes is small owing to the fact that the eigenvalue distribution of the Fisher information matrix of the NNs is strongly biased, which was shown by Takeishi et al. (2023) and has been refined in this paper. That is, we establish a tight upper bound on the risk of our MDL estimators. Note that our risk bound for the simple ReLU networks, of which the leading term is $O(d^2 \log n /n)$, is independent of the number of parameters $m$.

cs.IT↗

arXiv2023

Improved MDL Estimators Using Fiber Bundle of Local Exponential Families for Non-exponential Families

Minimum Description Length (MDL) estimators, using two-part codes for universal coding, are analyzed. For general parametric families under certain regularity conditions, we introduce a two-part code whose regret is close to the minimax regret, where regret of a code with respect to a target family M is the difference between the code length of the code and the ideal code length achieved by an element in M. This is a generalization of the result for exponential families by Grünwald. Our code is constructed by using an augmented structure of M with a bundle of local exponential families for data description, which is not needed for exponential families. This result gives a tight upper bound on risk and loss of the MDL estimators based on the theory introduced by Barron and Cover in 1991. Further, we show that we can apply the result to mixture families, which are a typical example of non-exponential families.

cs.IT↗

arXiv2023

Approximate Spectral Decomposition of Fisher Information Matrix for Simple ReLU Networks

We argue the Fisher information matrix (FIM) of one hidden layer networks with the ReLU activation function. For a network, let $W$ denote the $d \times p$ weight matrix from the $d$-dimensional input to the hidden layer consisting of $p$ neurons, and $v$ the $p$-dimensional weight vector from the hidden layer to the scalar output. We focus on the FIM of $v$, which we denote as $I$. Under certain conditions, we characterize the first three clusters of eigenvalues and eigenvectors of the FIM. Specifically, we show that 1) Since $I$ is non-negative owing to the ReLU, the first eigenvalue is the Perron-Frobenius eigenvalue. 2) For the cluster of the next maximum values, the eigenspace is spanned by the row vectors of $W$. 3) The direct sum of the eigenspace of the first eigenvalue and that of the third cluster is spanned by the set of all the vectors obtained as the Hadamard product of any pair of the row vectors of $W$. We confirmed by numerical calculation that the above is approximately correct when the number of hidden nodes is about 10000.

cs.LG↗

arXiv2018

An Improved Analysis of Least Squares Superposition Codes with Bernoulli Dictionary

For the additive white Gaussian noise channel with average power constraint, sparse superposition codes, proposed by Barron and Joseph in 2010, achieve the capacity. While the codewords of the original sparse superposition codes are made with a dictionary matrix drawn from a Gaussian distribution, we consider the case that it is drawn from a Bernoulli distribution. We show an improved upper bound on its block error probability with least squares decoding, which is fairly simplified and tighter bound than our previous result in 2014.

cs.IT↗

arXiv2016

A Note on Model Selection for Small Sample Regression

The risk estimator called "Direct Eigenvalue Estimator" (DEE) is studied. DEE was developed for small sample regression. In contrast to many existing model selection criteria, derivation of DEE requires neither any asymptotic assumption nor any prior knowledge about the noise variance and the noise distribution. It was reported that DEE performed well in small sample cases but DEE performed a little worse than the state-of-the-art ADJ. This seems somewhat counter-intuitive because DEE was developed for specifically regression problem by exploiting available information exhaustively, while ADJ was developed for general setting. In this paper, we point out that the derivation of DEE includes an inappropriate part in spite that the resultant form of DEE is valid in a sense. As its result, DEE cannot derive its potential. We introduce a class of valid risk estimators based on the idea of DEE and show that better risk estimators (mDEE) can be found in the class. By numerical experiments, we verify that mDEE often performs better than or at least equally the original DEE and ADJ.

stat.ME↗

arXiv2016

Minimum Description Length Principle in Supervised Learning with Application to Lasso

The minimum description length (MDL) principle in supervised learning is studied. One of the most important theories for the MDL principle is Barron and Cover's theory (BC theory), which gives a mathematical justification of the MDL principle. The original BC theory, however, can be applied to supervised learning only approximately and limitedly. Though Barron et al. recently succeeded in removing a similar approximation in case of unsupervised learning, their idea cannot be essentially applied to supervised learning in general. To overcome this issue, an extension of BC theory to supervised learning is proposed. The derived risk bound has several advantages inherited from the original BC theory. First, the risk bound holds for finite sample size. Second, it requires remarkably few assumptions. Third, the risk bound has a form of redundancy of the two-stage code for the MDL procedure. Hence, the proposed extension gives a mathematical justification of the MDL principle to supervised learning like the original BC theory. As an important example of application, new risk and (probabilistic) regret bounds of lasso with random design are derived. The derived risk bound holds for any finite sample size $n$ and feature number $p$ even if $n\ll p$ without boundedness of features in contrast to the past work. Behavior of the regret bound is investigated by numerical simulations. We believe that this is the first extension of BC theory to general supervised learning with random design without approximation.

cs.IT↗