Abstract--Federated learning is an important privacy-preserving multi-party learning paradigm, involving collaborative learning with others and local updating on private data. Model heterogeneity and catastrophic forgetting are two crucial challenges, which greatly limit the applicability and generalizability. This paper presents a novel FCCL+, federated correlation and similarity learning with nontarget distillation, facilitating the both intra-domain discriminability and inter-domain generalization. We construct cross-correlation matrix and align instance similarity distribution on both logits and feature levels, which effectively overcomes the communication barrier and improves the generalizable ability. For catastrophic forgetting in local updating stage, FCCL+ introduces Federated Non Target Distillation, which retains inter-domain knowledge while avoiding the optimization conflict issue, fulling distilling privileged inter-domain information through depicting posterior classes relation. Considering that there is no standard benchmark for evaluating existing heterogeneous federated learning under the same setting, we present a comprehensive benchmark with extensive representative methods under four domain shift scenarios, supporting both heterogeneous and homogeneous federated settings. Empirical results demonstrate the superiority of our method and the efficiency of modules on various scenarios.

Quantifying the dependence between high-dimensional random variables is central to statistical learning and inference. Two classical methods are canonical correlation analysis (CCA), which identifies maximally correlated projected versions of the original variables, and Shannon's mutual information, which is a universal dependence measure that also captures high-order dependencies. However, CCA only accounts for linear dependence, which may be insufficient for certain applications, while mutual information is often infeasible to compute/estimate in high dimensions. This work proposes a middle ground in the form of a scalable information-theoretic generalization of CCA, termed max-sliced mutual information (mSMI). mSMI equals the maximal mutual information between low-dimensional projections of the high-dimensional variables, which reduces back to CCA in the Gaussian case. It enjoys the best of both worlds: capturing intricate dependencies in the data while being amenable to fast computation and scalable estimation from samples. We show that mSMI retains favorable structural properties of Shannon's mutual information, like variational forms and identification of independence. We then study statistical estimation of mSMI, propose an efficiently computable neural estimator, and couple it with formal non-asymptotic error bounds. We present experiments that demonstrate the utility of mSMI for several tasks, encompassing independence testing, multi-view representation learning, algorithmic fairness, and generative modeling. We observe that mSMI consistently outperforms competing methods with little-to-no computational overhead.

Large Language Models (LLMs) have proven their exceptional capabilities in performing language-related tasks. However, their deployment poses significant challenges due to their considerable memory and storage requirements. In response to this issue, weight-only quantization, particularly 3 and 4-bit weight-only quantization, has emerged as one of the most viable solutions. As the number of bits decreases, the quantization grid broadens, thus emphasizing the importance of up and down rounding. While previous studies have demonstrated that fine-tuning up and down rounding with the addition of perturbations can enhance accuracy in some scenarios, our study is driven by the precise and limited boundary of these perturbations, where only the threshold for altering the rounding value is of significance. Consequently, we propose a concise and highly effective approach for optimizing the weight rounding task. Our method, named SignRound, involves lightweight block-wise tuning using signed gradient descent, enabling us to achieve outstanding results within 400 steps. SignRound competes impressively against recent methods without introducing additional inference overhead. The source code will be publicly available at \url{https://github.com/intel/neural-compressor} soon.

Spontaneous self-organization is ubiquitous in systems far from thermodynamic equilibrium. While organized structures that emerge dominate transport properties, universal representations that identify and describe these key objects remain elusive. Here, we introduce a theoretically-grounded framework for describing emergent organization that, via data-driven algorithms, is constructive in practice. Its building blocks are spacetime lightcones that embody how information propagates across a system through local interactions. We show that predictive equivalence classes of lightcones -- local causal states -- capture organized behaviors and coherent structures in complex spatiotemporal systems. Employing an unsupervised physics-informed machine learning algorithm and a high-performance computing implementation, we demonstrate automatically discovering coherent structures in two real world domain science problems. We show that local causal states identify vortices and track their power-law decay behavior in two-dimensional fluid turbulence. We then show how to detect and track familiar extreme weather events -- hurricanes and atmospheric rivers -- and discover other novel coherent structures associated with precipitation extremes in high-resolution climate data at the grid-cell level.

The soft Dice loss (SDL) has taken a pivotal role in numerous automated segmentation pipelines in the medical imaging community. Over the last years, some reasons behind its superior functioning have been uncovered and further optimizations have been explored. However, there is currently no implementation that supports its direct utilization in scenarios involving soft labels. Hence, a synergy between the use of SDL and research leveraging the use of soft labels, also in the context of model calibration, is still missing. In this work, we introduce Dice semimetric losses (DMLs), which (i) are by design identical to SDL in a standard setting with hard labels, but (ii) can be employed in settings with soft labels. Our experiments on the public QUBIQ, LiTS and KiTS benchmarks confirm the potential synergy of DMLs with soft labels (e.g.\ averaging, label smoothing, and knowledge distillation) over hard labels (e.g.\ majority voting and random selection). As a result, we obtain superior Dice scores and model calibration, which supports the wider adoption of DMLs in practice. The code is available at \href{https://github.com/zifuwanggg/JDTLosses}{https://github.com/zifuwanggg/JDTLosses}.

A typical machine learning (ML) development cycle for edge computing is to maximise the performance during model training and then minimise the memory/area footprint of the trained model for deployment on edge devices targeting CPUs, GPUs, microcontrollers, or custom hardware accelerators. This paper proposes a methodology for automatically generating predictor circuits for classification of tabular data with comparable prediction performance to conventional ML techniques while using substantially fewer hardware resources and power. The proposed methodology uses an evolutionary algorithm to search over the space of logic gates and automatically generates a classifier circuit with maximised training prediction accuracy. Classifier circuits are so tiny (i.e., consisting of no more than 300 logic gates) that they are called "Tiny Classifier" circuits, and can efficiently be implemented in ASIC or on an FPGA. We empirically evaluate the automatic Tiny Classifier circuit generation methodology or "Auto Tiny Classifiers" on a wide range of tabular datasets, and compare it against conventional ML techniques such as Amazon's AutoGluon, Google's TabNet and a neural search over Multi-Layer Perceptrons. Despite Tiny Classifiers being constrained to a few hundred logic gates, we observe no statistically significant difference in prediction performance in comparison to the best-performing ML baseline. When synthesised as a Silicon chip, Tiny Classifiers use 8-18x less area and 4-8x less power. When implemented as an ultra-low cost chip on a flexible substrate (i.e., FlexIC), they occupy 10-75x less area and consume 13-75x less power compared to the most hardware-efficient ML baseline. On an FPGA, Tiny Classifiers consume 3-11x fewer resources.

We propose a novel framework for incorporating unlabeled data into semi-supervised classification problems, where scenarios involving the minimization of either i) adversarially robust or ii) non-robust loss functions have been considered. Notably, we allow the unlabeled samples to deviate slightly (in total variation sense) from the in-domain distribution. The core idea behind our framework is to combine Distributionally Robust Optimization (DRO) with self-supervised training. As a result, we also leverage efficient polynomial-time algorithms for the training stage. From a theoretical standpoint, we apply our framework on the classification problem of a mixture of two Gaussians in $\mathbb{R}^d$, where in addition to the $m$ independent and labeled samples from the true distribution, a set of $n$ (usually with $n\gg m$) out of domain and unlabeled samples are gievn as well. Using only the labeled data, it is known that the generalization error can be bounded by $\propto\left(d/m\right)^{1/2}$. However, using our method on both isotropic and non-isotropic Gaussian mixture models, one can derive a new set of analytically explicit and non-asymptotic bounds which show substantial improvement on the generalization error compared ERM. Our results underscore two significant insights: 1) out-of-domain samples, even when unlabeled, can be harnessed to narrow the generalization gap, provided that the true data distribution adheres to a form of the "cluster assumption", and 2) the semi-supervised learning paradigm can be regarded as a special case of our framework when there are no distributional shifts. We validate our claims through experiments conducted on a variety of synthetic and real-world datasets.

Due to common architecture designs, symmetries exist extensively in contemporary neural networks. In this work, we unveil the importance of the loss function symmetries in affecting, if not deciding, the learning behavior of machine learning models. We prove that every mirror symmetry of the loss function leads to a structured constraint, which becomes a favored solution when either the weight decay or gradient noise is large. As direct corollaries, we show that rescaling symmetry leads to sparsity, rotation symmetry leads to low rankness, and permutation symmetry leads to homogeneous ensembling. Then, we show that the theoretical framework can explain the loss of plasticity and various collapse phenomena in neural networks and suggest how symmetries can be used to design algorithms to enforce hard constraints in a differentiable way.

We introduce a new tool, Transductive Local Rademacher Complexity (TLRC), to analyze the generalization performance of transductive learning methods and motivate new transductive learning algorithms. Our work extends the idea of the popular Local Rademacher Complexity (LRC) to the transductive setting with considerable changes compared to the analysis of typical LRC methods in the inductive setting. We present a localized version of Rademacher complexity based tool wihch can be applied to various transductive learning problems and gain sharp bounds under proper conditions. Similar to the development of LRC, we build TLRC by starting from a sharp concentration inequality for independent variables with variance information. The prediction function class of a transductive learning model is then divided into pieces with a sub-root function being the upper bound for the Rademacher complexity of each piece, and the variance of all the functions in each piece is limited. A carefully designed variance operator is used to ensure that the bound for the test loss on unlabeled test data in the transductive setting enjoys a remarkable similarity to that of the classical LRC bound in the inductive setting. We use the new TLRC tool to analyze the Transductive Kernel Learning (TKL) model, where the labels of test data are generated by a kernel function. The result of TKL lays the foundation for generalization bounds for two types of transductive learning tasks, Graph Transductive Learning (GTL) and Transductive Nonparametric Kernel Regression (TNKR). When the target function is low-dimensional or approximately low-dimensional, we design low rank methods for both GTL and TNKR, which enjoy particularly sharper generalization bounds by TLRC which cannot be achieved by existing learning theory methods, to the best of our knowledge.

Random feature model with a nonlinear activation function has been shown to perform asymptotically equivalent to a Gaussian model in terms of training and generalization errors. Analysis of the equivalent model reveals an important yet not fully understood role played by the activation function. To address this issue, we study the "parameters" of the equivalent model to achieve improved generalization performance for a given supervised learning problem. We show that acquired parameters from the Gaussian model enable us to define a set of optimal nonlinearities. We provide two example classes from this set, e.g., second-order polynomial and piecewise linear functions. These functions are optimized to improve generalization performance regardless of the actual form. We experiment with regression and classification problems, including synthetic and real (e.g., CIFAR10) data. Our numerical results validate that the optimized nonlinearities achieve better generalization performance than widely-used nonlinear functions such as ReLU. Furthermore, we illustrate that the proposed nonlinearities also mitigate the so-called double descent phenomenon, which is known as the non-monotonic generalization performance regarding the sample size and the model size.