Recent advancements in session-based recommendation models using deep learning techniques have demonstrated significant performance improvements. While they can enhance model sophistication and improve the relevance of recommendations, they also make it challenging to implement a scalable real-time solution. To addressing this challenge, we propose GRAINRec: a Graph and Attention Integrated session-based recommendation model that generates recommendations in real-time. Our scope of work is item recommendations in online retail where a session is defined as an ordered sequence of digital guest actions, such as page views or adds to cart. The proposed model generates recommendations by considering the importance of all items in the session together, letting us predict relevant recommendations dynamically as the session evolves. We also propose a heuristic approach to implement real-time inferencing that meets Target platform's service level agreement (SLA). The proposed architecture lets us predict relevant recommendations dynamically as the session evolves, rather than relying on pre-computed recommendations for each item. Evaluation results of the proposed model show an average improvement of 1.5% across all offline evaluation metrics. A/B tests done over a 2 week duration showed an increase of 10% in click through rate and 9% increase in attributable demand. Extensive ablation studies are also done to understand our model performance for different parameters.

Numerical simulation is a predominant tool for studying the dynamics in complex systems, but large-scale simulations are often intractable due to computational limitations. Here, we introduce the Neural Graph Simulator (NGS) for simulating time-invariant autonomous systems on graphs. Utilizing a graph neural network, the NGS provides a unified framework to simulate diverse dynamical systems with varying topologies and sizes without constraints on evaluation times through its non-uniform time step and autoregressive approach. The NGS offers significant advantages over numerical solvers by not requiring prior knowledge of governing equations and effectively handling noisy or missing data with a robust training scheme. It demonstrates superior computational efficiency over conventional methods, improving performance by over $10^5$ times in stiff problems. Furthermore, it is applied to real traffic data, forecasting traffic flow with state-of-the-art accuracy. The versatility of the NGS extends beyond the presented cases, offering numerous potential avenues for enhancement.

Ensuring fairness has emerged as one of the primary concerns in AI and its related algorithms. Over time, the field of machine learning fairness has evolved to address these issues. This paper provides an extensive overview of this field and introduces two formal frameworks to tackle open questions in machine learning fairness. In one framework, operator-valued optimisation and min-max objectives are employed to address unfairness in time-series problems. This approach showcases state-of-the-art performance on the notorious COMPAS benchmark dataset, demonstrating its effectiveness in real-world scenarios. In the second framework, the challenge of lacking sensitive attributes, such as gender and race, in commonly used datasets is addressed. This issue is particularly pressing because existing algorithms in this field predominantly rely on the availability or estimations of such attributes to assess and mitigate unfairness. Here, a framework for a group-blind bias-repair is introduced, aiming to mitigate bias without relying on sensitive attributes. The efficacy of this approach is showcased through analyses conducted on the Adult Census Income dataset. Additionally, detailed algorithmic analyses for both frameworks are provided, accompanied by convergence guarantees, ensuring the robustness and reliability of the proposed methodologies.

Interactive exploration of the unknown physical properties of objects such as stiffness, mass, center of mass, friction coefficient, and shape is crucial for autonomous robotic systems operating continuously in unstructured environments. Precise identification of these properties is essential to manipulate objects in a stable and controlled way, and is also required to anticipate the outcomes of (prehensile or non-prehensile) manipulation actions such as pushing, pulling, lifting, etc. Our study focuses on autonomously inferring the physical properties of a diverse set of various homogeneous, heterogeneous, and articulated objects utilizing a robotic system equipped with vision and tactile sensors. We propose a novel predictive perception framework for identifying object properties of the diverse objects by leveraging versatile exploratory actions: non-prehensile pushing and prehensile pulling. As part of the framework, we propose a novel active shape perception to seamlessly initiate exploration. Our innovative dual differentiable filtering with Graph Neural Networks learns the object-robot interaction and performs consistent inference of indirectly observable time-invariant object properties. In addition, we formulate a $N$-step information gain approach to actively select the most informative actions for efficient learning and inference. Extensive real-robot experiments with planar objects show that our predictive perception framework results in better performance than the state-of-the-art baseline and demonstrate our framework in three major applications for i) object tracking, ii) goal-driven task, and iii) change in environment detection.

As language models grow ever larger, so do their vocabularies. This has shifted the memory footprint of LLMs during training disproportionately to one single layer: the cross-entropy in the loss computation. Cross-entropy builds up a logit matrix with entries for each pair of input tokens and vocabulary items and, for small models, consumes an order of magnitude more memory than the rest of the LLM combined. We propose Cut Cross-Entropy (CCE), a method that computes the cross-entropy loss without materializing the logits for all tokens into global memory. Rather, CCE only computes the logit for the correct token and evaluates the log-sum-exp over all logits on the fly. We implement a custom kernel that performs the matrix multiplications and the log-sum-exp reduction over the vocabulary in flash memory, making global memory consumption for the cross-entropy computation negligible. This has a dramatic effect. Taking the Gemma 2 (2B) model as an example, CCE reduces the memory footprint of the loss computation from 24 GB to 1 MB, and the total training-time memory consumption of the classifier head from 28 GB to 1 GB. To improve the throughput of CCE, we leverage the inherent sparsity of softmax and propose to skip elements of the gradient computation that have a negligible (i.e., below numerical precision) contribution to the gradient. Experiments demonstrate that the dramatic reduction in memory consumption is accomplished without sacrificing training speed or convergence.

Traditional knowledge distillation focuses on aligning the student's predicted probabilities with both ground-truth labels and the teacher's predicted probabilities. However, the transition to predicted probabilities from logits would obscure certain indispensable information. To address this issue, it is intuitive to additionally introduce a logit-level loss function as a supplement to the widely used probability-level loss function, for exploiting the latent information of logits. Unfortunately, we empirically find that the amalgamation of the newly introduced logit-level loss and the previous probability-level loss will lead to performance degeneration, even trailing behind the performance of employing either loss in isolation. We attribute this phenomenon to the collapse of the classification head, which is verified by our theoretical analysis based on the neural collapse theory. Drawing from the theoretical analysis, we propose a novel method called dual-head knowledge distillation, which partitions the linear classifier into two classification heads responsible for different losses, thereby preserving the beneficial effects of both losses on the backbone while eliminating adverse influences on the classification head. Extensive experiments validate that our method can effectively exploit the information inside the logits and achieve superior performance against state-ofthe-art counterparts. Despite the remarkable success of deep neural networks (DNNs) in various fields, it is a significant challenge to deploy these large models in lightweight terminals (e.g., mobile phones), particularly under the constraint of computational resources or the requirement of short inference time. To mitigate this problem, knowledge distillation (KD) (Hinton et al., 2015) is widely investigated, which aims to improve the performance of a small network (a.k.a. the "student") by leveraging the expansive knowledge of a large network (a.k.a. the "teacher") to guide the training of the student network. Traditional KD techniques focus on minimizing the disparity in the predicted probabilities between the teacher and the student, which are typically the outputs of the softmax function. Nevertheless, the transformation from logits to predictive probabilities via the softmax function may lose some underlying information.

In the evolving field of digital pathology, Whole Slide Imaging (WSI) has emerged as a transformative technology, enabling the digitization of histopathological slides at gigapixel resolution. This advancement has not only facilitated remote diagnostics and educational opportunities but also opened new avenues for quantitative image analysis [1, 2]. Despite its potential, the sheer size and complexity of WSIs pose significant computational challenges, limiting the practicality of large-scale analysis and the application of advanced machine learning techniques [3, 4]. Whole slide imaging (WSI) represents a significant breakthrough in digital pathology, enabling the digitization of histological slides at high resolutions. This advancement allows for improved visualization, analysis, and management of tissue samples, essential for accurate disease diagnosis and research. However, the sheer size and complexity of WSIs pose unique challenges in image processing and analysis, necessitating innovative approaches for efficient and effective feature extraction and classification. Traditional methods for analyzing WSIs often rely on supervised learning techniques, which require extensive annotated datasets prepared by expert pathologists. This process is not only time-consuming but also prone to variability due to inter-observer differences.

Many studies have predicted SocioEconomic Position (SEP) for aggregated spatial units such as villages using satellite data, but SEP prediction at the household level and other sources of imagery have not been yet explored. We assembled a dataset of 975 households in a semi-rural district in southern Mozambique, consisting of self-reported asset, expenditure, and income SEP data, as well as multimodal imagery including satellite images and a ground-based photograph survey of 11 household elements. We fine-tuned a convolutional neural network to extract feature vectors from the images, which we then used in regression analyzes to model household SEP using different sets of image types. The best prediction performance was found when modeling asset-based SEP using random forest models with all image types, while the performance for expenditure- and income-based SEP was lower. Using SHAP, we observed clear differences between the images with the largest positive and negative effects, as well as identified the most relevant household elements in the predictions. Finally, we fitted an additional reduced model using only the identified relevant household elements, which had an only slightly lower performance compared to models using all images. Our results show how ground-based household photographs allow to zoom in from an area-level to an individual household prediction while minimizing the data collection effort by using explainable machine learning. The developed workflow can be potentially integrated into routine household surveys, where the collected household imagery could be used for other purposes, such as refined asset characterization and environmental exposure assessment.

In machine learning and data mining, outliers are data points that significantly differ from the dataset and often introduce irrelevant information that can induce bias in its statistics and models. Therefore, unsupervised methods are crucial to detect outliers if there is limited or no information about them. Global-Local Outlier Scores based on Hierarchies (GLOSH) is an unsupervised outlier detection method within HDBSCAN*, a state-of-the-art hierarchical clustering method. GLOSH estimates outlier scores for each data point by comparing its density to the highest density of the region they reside in the HDBSCAN* hierarchy. GLOSH may be sensitive to HDBSCAN*'s minpts parameter that influences density estimation. With limited knowledge about the data, choosing an appropriate minpts value beforehand is challenging as one or some minpts values may better represent the underlying cluster structure than others. Additionally, in the process of searching for ``potential outliers'', one has to define the number of outliers n a dataset has, which may be impractical and is often unknown. In this paper, we propose an unsupervised strategy to find the ``best'' minpts value, leveraging the range of GLOSH scores across minpts values to identify the value for which GLOSH scores can best identify outliers from the rest of the dataset. Moreover, we propose an unsupervised strategy to estimate a threshold for classifying points into inliers and (potential) outliers without the need to pre-define any value. Our experiments show that our strategies can automatically find the minpts value and threshold that yield the best or near best outlier detection results using GLOSH.

Humans naturally follow distinct patterns when conducting their daily activities, which are driven by established practices and processes, such as production workflows, social norms and daily routines. Human activity recognition (HAR) algorithms usually use neural networks or machine learning techniques to analyse inherent relationships within the data. However, these approaches often overlook the contextual information in which the data are generated, potentially limiting their effectiveness. We propose a novel approach that incorporates process information from context to enhance the HAR performance. Specifically, we align probabilistic events generated by machine learning models with process models derived from contextual information. This alignment adaptively weighs these two sources of information to optimise HAR accuracy. Our experiments demonstrate that our approach achieves better accuracy and Macro F1-score compared to baseline models.