Diffusion Language Models (DLMs) offer a compelling alternative to autoregressive (AR) generation by enabling bidirectional context and iterative refinement. However, their reliability under natural input noise and adversarial attacks remains under-explored. To address this, we systematically evaluate DLM robustness and calibration against AR baselines, using two parameter-matched pairs (LLaDA-8B vs. LLaMA-3-8B and Dream-7B vs. Qwen2.5-7B) across 32 natural perturbation conditions, adversarial gradient probes, and mechanistic hidden-state analyses. This paired design effectively isolates architecture-intrinsic properties from weight-dependent behaviors. We find a nuanced robustness profile: while highly stochastic DLM loss landscapes naturally resist gradient-based adversarial suffixes, they provide no guaranteed defense against natural noise, proving that everyday robustness is weight-dependent rather than inherently architectural. Furthermore, DLMs exhibit systematic overconfidence, presenting a practical deployment hazard. Most crucially, mechanistic probing reveals that all models perfectly encode input corruption, isolating behavioral fragility entirely to a decoder routing failure. Consistent with this diagnosis, we show that surface-level prompt patching fails to improve over noisy baselines. Ultimately, DLM robustness cannot be patched on; it must be fundamentally integrated into the iterative decoding loop.

Diffusion language models use broad context to create text, suggesting they might handle input noise better than standard models. Testing reveals this is only partially true. Internally, diffusion models detect text errors highly accurately. Externally, their reported certainty ignores this signal. As accuracy drops due to noise, confidence stays near its maximum and the ability to correctly rank answers degrades toward random chance. We call this mismatch the representation confidence gap. The visible concentration of high certainty scores is a misleading surface symptom. Standard math adjustments remove this concentration but fail to fix the underlying loss of ranking order. This ranking deficit favors standard models under noisy conditions and resists common remedies. Matching training recovers accuracy but not ranking, while score recalibration and input level error signals cannot reorder the final answers. However, the information needed to properly evaluate an answer survives in the hidden states. A lightweight extraction tool uses this signal to improve ranking. This approach is highly efficient because it leaves the base model completely frozen and requires zero additional text generation steps. We present this tool to prove the signal exists, while clearly noting its limits. Ultimately, certainty reliability is a more pressing limit than overall accuracy under noisy conditions.

Sequence modeling is a crucial area across various domains, including Natural Language Processing (NLP), speech recognition, time series forecasting, music generation, and bioinformatics. Recurrent Neural Networks (RNNs) and Long Short Term Memory Networks (LSTMs) have historically dominated sequence modeling tasks like Machine Translation, Named Entity Recognition (NER), etc. However, the advancement of transformers has led to a shift in this paradigm, given their superior performance. Yet, transformers suffer from O(N2) attention complexity and challenges in handling inductive bias. Several variations have been proposed to address these issues which use spectral networks or convolutions and have performed well on a range of tasks. However, they continue to struggle with managing long sequences. State Space Models(SSMs) have emerged as promising alternatives for sequence modeling paradigms in existing state space models. In this survey, we categorize the foundational SSMs based on three paradigms namely, Gating architectures, Structural architectures, and Recurrent architectures. This survey also highlights diverse applications of SSMs across domains such as vision, video, audio, speech, language (especially long sequence modeling), medical (including genomics), chemical (like drug design), recommendation systems, and time series analysis, including tabular data. Moreover, we consolidate the performance of SSMs on benchmark datasets like Long Range Arena (LRA), Text datasets, Image datasets, as well as video datasets and various time series datasets. The project page for Mamba-360 work is available on this webpage.https://github.com/badripatro/mamba360.

Vision transformers have been applied successfully for image recognition tasks. There have been either multi-headed self-attention based (ViT \cite{dosovitskiy2020image}, DeIT, \cite{touvron2021training}) similar to the original work in textual models or more recently based on spectral layers (Fnet\cite{lee2021fnet}, GFNet\cite{rao2021global}, AFNO\cite{guibas2021efficient}). We hypothesize that both spectral and multi-headed attention plays a major role. We investigate this hypothesis through this work and observe that indeed combining spectral and multi-headed attention layers provides a better transformer architecture. We thus propose the novel Spectformer architecture for transformers that combines spectral and multi-headed attention layers. We believe that the resulting representation allows the transformer to capture the feature representation appropriately and it yields improved performance over other transformer representations. For instance, it improves the top-1 accuracy by 2\% on ImageNet compared to both GFNet-H and LiT. SpectFormer-S reaches 84.25\% top-1 accuracy on ImageNet-1K (state of the art for small version). Further, Spectformer-L achieves 85.7\% that is the state of the art for the comparable base version of the transformers. We further ensure that we obtain reasonable results in other scenarios such as transfer learning on standard datasets such as CIFAR-10, CIFAR-100, Oxford-IIIT-flower, and Standford Car datasets. We then investigate its use in downstream tasks such of object detection and instance segmentation on the MS-COCO dataset and observe that Spectformer shows consistent performance that is comparable to the best backbones and can be further optimized and improved. Hence, we believe that combined spectral and attention layers are what are needed for vision transformers.

We investigate the question of whether transformers are effective for long-term time series forecasting. They suffer from temporal information loss, even though they use positional encoding to embed order information. This is due to the inherent permutation-invariant nature of attention nets. Mamba has a stability issue at scale, so we use a novel EinFFT method for getting positive Eigenvalues with the help of fast Fourier transforms. We introduce SiMBA-TS, a novel approach combining the selective scan SSM called Mamba for token mixing and a new method called EinFFT for channel mixing. Our benchmark studies on seven-time series datasets demonstrate that SiMBA-TS achieves state-of-the-art performance, outperforming transformers such as PatchTST, TimeNet, and CrossFormer. Additionally, our evaluation of a proprietary industrial dataset confirms that SiMBA-TS is the leading model for time series forecasting.

We have seen recent advances in the fields of Machine Learning (ML), Deep Learning (DL), and Artificial intelligence (AI) that the models are becoming increasingly complex and large in terms of architecture and parameter size. These complex ML/DL models have beaten the state of the art in most fields of computer science like computer vision, NLP, tabular data prediction and time series forecasting, etc. With the increase in models’ performance, model explainability and interpretability has become essential to explain/justify model outcome, especially for business use cases. There has been significant improvement in the domain of model explainability for Computer Vision and Natural Language Processing (NLP) tasks with fundamental research for both black-box and white-box techniques. In this paper, we proposed novel time series explainability techniques SPASE for black-box time series model forecasting and anomaly detection problems.

Azure Core workload insights have time-series data with different metric units. Faults or Anomalies are observed in these time-series data owing to faults observed with respect to metric name, resources region, dimensions, and its dimension value associated with the data. For Azure Core, an important task is to highlight faults or anomalies to the user on a dashboard that they can perceive easily. The number of anomalies reported should be highly significant and in a limited number, e.g., 5-20 anomalies reported per hour. The reported anomalies will have significant user perception and high reconstruction error in any time-series forecasting model. Hence, our task is to automatically identify 'high significant anomalies' and their associated information for user perception.

Transformers have revolutionized image modeling tasks with adaptations like DeIT, Swin, SVT, Biformer, STVit, and FDVIT. However, these models often face challenges with inductive bias and high quadratic complexity, making them less efficient for high-resolution images. State space models (SSMs) such as Mamba, V-Mamba, ViM, and SiMBA offer an alternative to handle high resolution images in computer vision tasks. These SSMs encounter two major issues. First, they become unstable when scaled to large network sizes. Second, although they efficiently capture global information in images, they inherently struggle with handling local information. To address these challenges, we introduce Heracles, a novel SSM that integrates a local SSM, a global SSM, and an attention-based token interaction module. Heracles leverages a Hartely kernel-based state space model for global image information, a localized convolutional network for local details, and attention mechanisms in deeper layers for token interactions. Our extensive experiments demonstrate that Heracles-C-small achieves state-of-the-art performance on the ImageNet dataset with 84.5% top-1 accuracy. Heracles-C-Large and Heracles-C-Huge further improve accuracy to 85.9% and 86.4%, respectively. Additionally, Heracles excels in transfer learning tasks on datasets such as CIFAR-10, CIFAR-100, Oxford Flowers, and Stanford Cars, and in instance segmentation on the MSCOCO dataset. Heracles also proves its versatility by achieving state-of-the-art results on seven time-series datasets, showcasing its ability to generalize across domains with spectral data, capturing both local and global information. The project page is available at this link.

Download This Paper Open PDF in Browser Add Paper to My Library Share: Permalink Using these links will ensure access to this page indefinitely Copy URL Copy DOI

Transformers used in vision have been investigated through diverse architectures - ViT, PVT, and Swin. These have worked to improve the attention mechanism and make it more efficient. Differently, the need for including local information was felt, leading to incorporating convolutions in transformers such as CPVT and CvT. Global information is captured using a complex Fourier basis to achieve global token mixing through various methods, such as AFNO, GFNet, and Spectformer. We advocate combining three diverse views of data - local, global, and long-range dependence. We also investigate the simplest global representation using only the real domain spectral representation - obtained through the Hartley transform. We use a convolutional operator in the initial layers to capture local information. Through these two contributions, we are able to optimize and obtain a spectral convolution transformer (SCT) that provides improved performance over the state-of-the-art methods while reducing the number of parameters. Through extensive experiments, we show that SCT-C-small gives state-of-the-art performance on the ImageNet dataset and reaches 84.5\% top-1 accuracy, while SCT-C-Large reaches 85.9\% and SCT-C-Huge reaches 86.4\%. We evaluate SCT on transfer learning on datasets such as CIFAR-10, CIFAR-100, Oxford Flower, and Stanford Car. We also evaluate SCT on downstream tasks i.e. instance segmentation on the MSCOCO dataset. The project page is available on this webpage.\url{https://github.com/badripatro/sct}

Transformers have widely adopted attention networks for sequence mixing and MLPs for channel mixing, playing a pivotal role in achieving breakthroughs across domains. However, recent literature highlights issues with attention networks, including low inductive bias and quadratic complexity concerning input sequence length. State Space Models (SSMs) like S4 and others (Hippo, Global Convolutions, liquid S4, LRU, Mega, and Mamba), have emerged to address the above issues to help handle longer sequence lengths. Mamba, while being the state-of-the-art SSM, has a stability issue when scaled to large networks for computer vision datasets. We propose SiMBA, a new architecture that introduces Einstein FFT (EinFFT) for channel modeling by specific eigenvalue computations and uses the Mamba block for sequence modeling. Extensive performance studies across image and time-series benchmarks demonstrate that SiMBA outperforms existing SSMs, bridging the performance gap with state-of-the-art transformers. Notably, SiMBA establishes itself as the new state-of-the-art SSM on ImageNet and transfer learning benchmarks such as Stanford Car and Flower as well as task learning benchmarks as well as seven time series benchmark datasets. The project page is available on this website ~\url{https://github.com/badripatro/Simba}.

Abstract We have seen recent advances in the field of Machine Learning (ML), Deep Learning (DL), and Artificial intelligence (AI) that the models are becoming increasingly complex and large in terms of architecture and parameter size. These complex ML/DL models have beaten the state of the art in most fields of computer science like computer vision, NLP, tabular data prediction and time series forecasting, etc. With the increase in models’ performance, model explainability and interpretability has become essential to explain/justify model outcome, especially for business use cases. There has been significant improvement in the domain of model explainability for Computer Vision and Natural Language Processing (NLP) tasks with fundamental research for both black-box and white-box techniques. In this paper, we proposed novel time series explainability techniques SPASE for black-box time series model forecasting and anomaly detection problems. We also outline details of its implementation and integration into Microsoft Azure products.

Transformers are widely used for solving tasks in natural language processing, computer vision, speech, and music domains. In this paper, we talk about the efficiency of transformers in terms of memory (the number of parameters), computation cost (number of floating points operations), and performance of models, including accuracy, the robustness of the model, and fair \& bias-free features. We mainly discuss the vision transformer for the image classification task. Our contribution is to introduce an efficient 360 framework, which includes various aspects of the vision transformer, to make it more efficient for industrial applications. By considering those applications, we categorize them into multiple dimensions such as privacy, robustness, transparency, fairness, inclusiveness, continual learning, probabilistic models, approximation, computational complexity, and spectral complexity. We compare various vision transformer models based on their performance, the number of parameters, and the number of floating point operations (FLOPs) on multiple datasets.

Vision transformers have gained significant attention and achieved state-of-the-art performance in various computer vision tasks, including image classification, instance segmentation, and object detection. However, challenges remain in addressing attention complexity and effectively capturing fine-grained information within images. Existing solutions often resort to down-sampling operations, such as pooling, to reduce computational cost. Unfortunately, such operations are non-invertible and can result in information loss. In this paper, we present a novel approach called Scattering Vision Transformer (SVT) to tackle these challenges. SVT incorporates a spectrally scattering network that enables the capture of intricate image details. SVT overcomes the invertibility issue associated with down-sampling operations by separating low-frequency and high-frequency components. Furthermore, SVT introduces a unique spectral gating network utilizing Einstein multiplication for token and channel mixing, effectively reducing complexity. We show that SVT achieves state-of-the-art performance on the ImageNet dataset with a significant reduction in a number of parameters and FLOPS. SVT shows 2\% improvement over LiTv2 and iFormer. SVT-H-S reaches 84.2\% top-1 accuracy, while SVT-H-B reaches 85.2\% (state-of-art for base versions) and SVT-H-L reaches 85.7\% (again state-of-art for large versions). SVT also shows comparable results in other vision tasks such as instance segmentation. SVT also outperforms other transformers in transfer learning on standard datasets such as CIFAR10, CIFAR100, Oxford Flower, and Stanford Car datasets. The project page is available on this webpage.\url{https://badripatro.github.io/svt/}.

Scene Text Recognition (STR) enables processing and understanding of the text in the wild. However, roadblocks like natural degradation, blur, and uneven lighting in the captured images result in poor accuracy during detection and recognition. Previous approaches have introduced Super-Resolution (SR) as a processing step between detection and recognition; however, post enhancement, there is a significant drop in the quality of the reconstructed text in the image. This drop is especially significant in the healthcare domain because any loss in accuracy can be detrimental. This paper will quantitatively show the drop in quality of the text in an image from the existing SR techniques across multiple optimization-based and GAN-based models. We propose a new loss function for training and an improved deep neural network architecture to address these shortcomings and recover text with sharp boundaries in the SR images. We also show that the Peak Signal-to-Noise Ratio (PSNR) and the Structural Similarity Index Measure (SSIM) scores are not effective metrics for identifying the quality of the text in an SR image. Extensive experiments show that our model achieves better accuracy and visual improvements against state-of-the-art methods in terms of text recognition accuracy. We plan to add our module on SR in the near future to our already deployed solution for text extraction from product images for our company.

Missing values, widely called as \textit{sparsity} in literature, is a common characteristic of many real-world datasets. Many imputation methods have been proposed to address this problem of data incompleteness or sparsity. However, the accuracy of a data imputation method for a given feature or a set of features in a dataset is highly dependent on the distribution of the feature values and its correlation with other features. Another problem that plagues industry deployments of machine learning (ML) solutions is concept drift detection, which becomes more challenging in the presence of missing values. Although data imputation and concept drift detection have been studied extensively, little work has attempted a combined study of the two phenomena, i.e., concept drift detection in the presence of sparsity. In this work, we carry out a systematic study of the following: (i) different patterns of missing values, (ii) various statistical and ML based data imputation methods for different kinds of sparsity, (iii) several concept drift detection methods, (iv) practical analysis of the various drift detection metrics, (v) selecting the best concept drift detector given a dataset with missing values based on the different metrics. We first analyze it on synthetic data and publicly available datasets, and finally extend the findings to our deployed solution of automated change risk assessment system. One of the major findings from our empirical study is the absence of supremacy of any one concept drift detection method across all the relevant metrics. Therefore, we adopt a majority voting based ensemble of concept drift detectors for abrupt and gradual concept drifts. Our experiments show optimal or near optimal performance can be achieved for this ensemble method across all the metrics.

Reducing the number of failures in a production system is one of the most challenging problems in technology driven industries, such as, the online retail industry. To address this challenge, change management has emerged as a promising sub-field in operations that manages and reviews the changes to be deployed in production in a systematic manner. However, it is practically impossible to manually review a large number of changes on a daily basis and assess the risk associated with them. This warrants the development of an automated system to assess the risk associated with a large number of changes. There are a few commercial solutions available to address this problem but those solutions lack the ability to incorporate domain knowledge and continuous feedback from domain experts into the risk assessment process. As part of this work, we aim to bridge the gap between model-driven risk assessment of change requests and the assessment of domain experts by building a continuous feedback loop into the risk assessment process. Here we present our work to build an end-to-end machine learning system along with the discussion of some of practical challenges we faced related to extreme skewness in class distribution, concept drift, estimation of the uncertainty associated with the model's prediction and the overall scalability of the system.

Extracting texts of various shapes, sizes, and orientations from images containing multiple objects is an important problem in many contexts, especially, in connection to E-commerce. In the context of the scale at which Walmart operates, the text from an image can be a richer and more accurate source of data than human inputs and can be used in several applications such as Attribute Extraction, Offensive Text Classification, Product Matching among others. The motivation of this particular work has come from different business requirements such as flagging products whose images contain words that are non-compliant with organizational policies and building an efficient automated system to identify similar products by comparing the information contained in their respective product images and many others. Existing methods fail to address domain specific challenges like high entropy, different orientations, and small texts in product images adequately. In this work, we provide a solution that not only addresses these challenges but is also proven to work at a million image scale for various retail business units within Walmart. Extensive experimentation revealed that our proposed solution has been able to save around 30% computational cost in both the training and the inference stages.

The importance of usage of massive datasets in machine learning is continuously growing. Such massive datasets pose serious challenge to robustness of the model to label-noise which is a critical property for any classifier. This challenge primarily stems from the fact that such datasets often contain biases which unfairly disadvantage certain groups. Non-random label-noise can be introduced into the data because of automatic labeling, lack of domain expertise while manual labeling and data poisoning attack by adversaries. Classifiers trained on such datasets can inherit these biases. In most of the prior work it is assumed that the entire training data are potentially corrupted. However it is always possible to curate small amount of trusted data for testing and validation and a part of it can be used up for denoising training data. Recently a methodology has been proposed to de-noise label-corrupted data using a sample of trusted data. The core of the methodology is based on the assumption that adversarial process has no access to true labels of the data. This is quite strong an assumption as it underestimates the capacity of the adversarial process. As part of this work, we proposed a more generic framework to correct label noise without any assumption on the access of the adversarial process to true labels of the data. Experimental results suggest performance improvement by our method over existing methods.

Our data scientists are adept in using machine learning algorithms and building model out of it, and they are at ease with their local machine to do them. But, when it comes to building the same model from the platform, they find it slightly challenging and need assistance from the platform team. Based on the survey results, the major challenge was platform complexity, but it is hard to deduce actionable items or accurate details to make the system simple. The complexity feedback was very generic, so we decided to break it down into two logical challenges: Education & Training and Simplicity-of-Platform. We have developed a system to find these two challenges in our platform, which we call an Analyzer. In this paper, we explain how it was built and it’s impact on the evolution of our machine learning platform. Our work aims to address these challenges and provide guidelines on how to empower machine learning platform team to know the data scientist’s bottleneck in building model.