With the increasing complexity of generative AI models, post-training quantization (PTQ) has emerged as a promising solution for deploying hyper-scale models on edge devices such as mobile and TVs.Existing PTQ schemes, however, consume considerable time and resources, which could be a bottleneck in real situations where frequent model updates and multiple hyperparameter tunings are required.As a cost-effective alternative, learning-free PTQ schemes have been proposed. However, the performance is somewhat limited because they cannot consider the inter-layer dependency within the attention module, which is a significant feature of Transformers.In this paper, we thus propose a novel PTQ algorithm that balances accuracy and efficiency.The key idea of the proposed algorithm called aespa is to perform quantization layer-wise for efficiency while targeting attention-wise reconstruction to consider the cross-layer dependency.Through extensive experiments on various language models and complexity analysis, we demonstrate that aespa is accurate and efficient in quantizing Transformer models.

Quantization is a fundamental optimization for many machine learning (ML) use cases, including compressing gradients, model weights and activations, and datasets. The most accurate form of quantization is adaptive, where the error is minimized with respect to a given input rather than optimizing for the worst case. However, optimal adaptive quantization methods are considered infeasible in terms of both their runtime and memory requirements.We revisit the Adaptive Stochastic Quantization (ASQ) problem and present algorithms that find optimal solutions with asymptotically improved time and space complexities. Our experiments indicate that our algorithms may open the door to using ASQ more extensively in a variety of ML applications. We also present an even faster approximation algorithm for quantizing large inputs on the fly.

Multi-distribution or collaborative learning involves learning a single predictor that works well across multiple data distributions, using samples from each during training. Recent research on multi-distribution learning, focusing on binary loss and finite VC dimension classes, has shown near-optimal sample complexity that is achieved with oracle efficient algorithms. That is, these algorithms are computationally efficient given an efficient ERM for the class. Unlike in classical PAC learning, where the optimal sample complexity is achieved with deterministic predictors, current multi-distribution learning algorithms output randomized predictors. This raises the question: can these algorithms be derandomized to produce a deterministic predictor for multiple distributions? Through a reduction to discrepancy minimization, we show that derandomizing multi-distribution learning is computationally hard, even when ERM is computationally efficient. On the positive side, we identify a structural condition enabling an efficient black-box reduction, converting existing randomized multi-distribution predictors into deterministic ones.

Compared to existing toolkits: (i) We support NLP and Computer Vision classification as well as standard tabular problems.

We study the generalization of two-layer ReLU neural networks in a univariate nonparametric regression problem with noisy labels. This is a problem where kernels (\emph{e.g.} NTK) are provably sub-optimal and benign overfitting does not happen, thus disqualifying existing theory for interpolating (0-loss, global optimal) solutions. We present a new theory of generalization for local minima that gradient descent with a constant learning rate can \emph{stably} converge to. We show that gradient descent with a fixed learning rate $\eta$ can only find local minima that represent smooth functions with a certain weighted \emph{first order total variation} bounded by $1/\eta - 1/2 + \widetilde{O}(\sigma + \sqrt{\mathrm{MSE}})$ where $\sigma$ is the label noise level, $\mathrm{MSE}$ is short for mean squared error against the ground truth, and $\widetilde{O}(\cdot)$ hides a logarithmic factor. Under mild assumptions, we also prove a nearly-optimal MSE bound of $\widetilde{O}(n^{-4/5})$ within the strict interior of the support of the $n$ data points. Our theoretical results are validated by extensive simulation that demonstrates large learning rate training induces sparse linear spline fits. To the best of our knowledge, we are the first to obtain generalization bound via minima stability in the non-interpolation case and the first to show ReLU NNs without regularization can achieve near-optimal rates in nonparametric regression.

We propose adaptive, line-search-free second-order methods with optimal rate of convergence for solving convex-concave min-max problems. By means of an adaptive step size, our algorithms feature a simple update rule that requires solving only one linear system per iteration, eliminating the need for line-search or backtracking mechanisms. Specifically, we base our algorithms on the optimistic method and appropriately combine it with second-order information. Moreover, distinct from common adaptive schemes, we define the step size recursively as a function of the gradient norm and the prediction error in the optimistic update. We first analyze a variant where the step size requires knowledge of the Lipschitz constant of the Hessian. Under the additional assumption of Lipschitz continuous gradients, we further design a parameter-free version by tracking the Hessian Lipschitz constant locally and ensuring the iterates remain bounded. We also evaluate the practical performance of our algorithm by comparing it to existing second-order algorithms for minimax optimization.

We present a novel bi-directional Transformer architecture (BiXT) which scales linearly with input size in terms of computational cost and memory consumption, but does not suffer the drop in performance or limitation to only one input modality seen with other efficient Transformer-based approaches. BiXT is inspired by the Perceiver architectures but replaces iterative attention with an efficient bi-directional cross-attention module in which input tokens and latent variables attend to each other simultaneously, leveraging a naturally emerging attention-symmetry between the two. This approach unlocks a key bottleneck experienced by Perceiver-like architectures and enables the processing and interpretation of both semantics ('what') and location ('where') to develop alongside each other over multiple layers -- allowing its direct application to dense and instance-based tasks alike. By combining efficiency with the generality and performance of a full Transformer architecture, BiXT can process longer sequences like point clouds, text or images at higher feature resolutions and achieves competitive performance across a range of tasks like point cloud part segmentation, semantic image segmentation, image classification, hierarchical sequence modeling and document retrieval. Our experiments demonstrate that BiXT models outperform larger competitors by leveraging longer sequences more efficiently on vision tasks like classification and segmentation, and perform on par with full Transformer variants on sequence modeling and document retrieval -- but require 28\% fewer FLOPs and are up to $8.4\times$ faster.

However, it exhibits the low expert activation issue, i.e., only a small subset of experts are activated for optimization, leading to suboptimal performance and limiting its effectiveness in learning a larger number of experts in complex tasks. In this paper, we propose Multi-Head Mixture-of-Experts (MH-MoE). MH-MoE split each input token into multiple sub-tokens, then these sub-tokens are assigned to and processed by a diverse set of experts in parallel, and seamlessly reintegrated into the original token form. The above operations enables MH-MoE to significantly enhance expert activation while collectively attend to information from various representation spaces within different experts to deepen context understanding. Besides, it's worth noting that our MH-MoE is straightforward to implement and decouples from other SMoE frameworks, making it easy to integrate with these frameworks for enhanced performance. Extensive experimental results across different parameter scales (300M to 7B) and three pre-training tasks--English-focused language modeling, multi-lingual language modeling and masked multi-modality modeling--along with multiple downstream validation tasks, demonstrate the effectiveness of MH-MoE.

The emerging vision foundation model (VFM) has inherited the ability to generalize to unseen images.Nevertheless, the key challenge of domain-generalized semantic segmentation (DGSS) lies in the domain gap attributed to the cross-domain styles, i.e., the variance of urban landscape and environment dependencies.Hence, maintaining the style-invariant property with varying domain styles becomes the key bottleneck in harnessing VFM for DGSS. The frequency space after Haar wavelet transformation provides a feasible way to decouple the style information from the domain-invariant content, since the content and style information are retained in the low-and high-frequency components of the space, respectively. To this end, we propose a novel Frequency-Adapted (FADA) learning scheme to advance the frontier.Its overall idea is to separately tackle the content and style information by frequency tokens throughout the learning process.Particularly, the proposed FADA consists of two branches, i.e., low-and high-frequency branches. The former one is able to stabilize the scene content, while the latter one learns the scene styles and eliminates its impact to DGSS. Experiments conducted on various DGSS settings show the state-of-the-art performance of our FADA and its versatility to a variety of VFMs.Source code is available at \url{https://github.com/BiQiWHU/FADA}.

Message Passing Graph Neural Networks (MPGNNs) have emerged as the preferred method for modeling complex interactions across diverse graph entities. While the theory of such models is well understood, their aggregation module has not received sufficient attention. Sum-based aggregators have solid theoretical foundations regarding their separation capabilities. However, practitioners often prefer using more complex aggregations and mixtures of diverse aggregations. In this work, we unveil a possible explanation for this gap. We claim that sum-based aggregators fail to mix features belonging to distinct neighbors, preventing them from succeeding at downstream tasks.To this end, we introduce Sequential Signal Mixing Aggregation (SSMA), a novel plug-and-play aggregation for MPGNNs. SSMA treats the neighbor features as 2D discrete signals and sequentially convolves them, inherently enhancing the ability to mix features attributed to distinct neighbors. By performing extensive experiments, we show that when combining SSMA with well-established MPGNN architectures, we achieve substantial performance gains across various benchmarks, achieving new state-of-the-art results in many settings.We published our code at https://almogdavid.github.io/SSMA/.