We present a new approach to understanding the relationship between loss curvature and input-output model behaviour in deep learning. Specifically, we use existing empirical analyses of the spectrum of deep network loss Hessians to ground an ansatz tying together the loss Hessian and the input-output Jacobian over training samples during the training of deep neural networks. We then prove a series of theoretical results which quantify the degree to which the input-output Jacobian of a model approximates its Lipschitz norm over a data distribution, and deduce a novel generalisation bound in terms of the empirical Jacobian. We use our ansatz, together with our theoretical results, to give a new account of the recently observed progressive sharpening phenomenon, as well as the generalisation properties of flat minima. Experimental evidence is provided to validate our claims.

Dana Goodyear's article about the scientist He Jiankui captured the terrifying potential of gene editing ("Dangerous Designs," September 11th). However, many of the problems that certain scientists are trying to remedy with the gene-editing tool CRISPR already have controversy-free solutions. Conditions such as Batten disease, caused when a baby inherits one pathogenic gene from each parent, can be prevented by pre-conception screenings of would-be parents. Hypertrophic cardiomyopathy, which is inherited from one parent, can be averted using pre-implant genetic diagnosis, a common I.V.F. Gene editing will eventually have a place in clinical medicine, but its use will likely be minimal, compared with these currently accessible, effective, and less buzzworthy techniques.

This research aims to evaluate the performance of the rotors and study the behavior of the human driver in interacting with the rotors. In recent years, rotors have been increasingly used between countries due to their safety, capacity, and environmental advantages, and because they provide safe and fluid flows of vehicles for transit and integration. It turns out that roundabouts can significantly reduce speed at twisting intersections, entry speed and the resulting effect on speed depends on the rating of road users. In our research, (bus, car, truck) drivers were given special attention and their behavior was categorized into (conservative, normal, aggressive). Anticipating and recognizing driver behavior is an important challenge. Therefore, the aim of this research is to study the effect of roundabouts on these classifiers and to develop a method for predicting the behavior of road users at roundabout intersections. Safety is primarily due to two inherent features of the rotor. First, by comparing the data collected and processed in order to classify and evaluate drivers' behavior, and comparing the speeds of the drivers (bus, car and truck), the speed of motorists at crossing the roundabout was more fit than that of buses and trucks. We looked because the car is smaller and all parts of the rotor are visible to it. So drivers coming from all directions have to slow down, giving them more time to react and mitigating the consequences in the event of an accident. Second, with fewer conflicting flows (and points of conflict), drivers only need to look to their left (in right-hand traffic) for other vehicles, making their job of crossing the roundabout easier as there is less need to split attention between different directions.

Employers are adopting algorithmic hiring technology throughout the recruitment pipeline. Algorithmic fairness is especially applicable in this domain due to its high stakes and structural inequalities. Unfortunately, most work in this space provides partial treatment, often constrained by two competing narratives, optimistically focused on replacing biased recruiter decisions or pessimistically pointing to the automation of discrimination. Whether, and more importantly what types of, algorithmic hiring can be less biased and more beneficial to society than low-tech alternatives currently remains unanswered, to the detriment of trustworthiness. This multidisciplinary survey caters to practitioners and researchers with a balanced and integrated coverage of systems, biases, measures, mitigation strategies, datasets, and legal aspects of algorithmic hiring and fairness. Our work supports a contextualized understanding and governance of this technology by highlighting current opportunities and limitations, providing recommendations for future work to ensure shared benefits for all stakeholders.

In this article we investigate smoothing (i.e., optimisation-based) estimation techniques for robot localization using an IMU aided by other localization sensors. We more particularly focus on Invariant Smoothing (IS), a variant based on the use of nontrivial Lie groups from robotics. We study the recently introduced Two Frames Group (TFG), and prove it can fit into the framework of Invariant Smoothing in order to better take into account the IMU biases, as compared to the state-of-the-art in robotics. Experiments based on the KITTI dataset show the proposed framework compares favorably to the state-of-the-art smoothing methods in terms of robustness in some challenging situations.

Recommendation foundation model utilizes large language models (LLM) for recommendation by converting recommendation tasks into natural language tasks. It enables generative recommendation which directly generates the item(s) to recommend rather than calculating a ranking score for each and every candidate item as in traditional recommendation models, simplifying the recommendation pipeline from multi-stage filtering to single-stage filtering. To avoid generating excessively long text and hallucinated recommendations when deciding which item(s) to recommend, creating LLM-compatible item IDs to uniquely identify each item is essential for recommendation foundation models. In this study, we systematically examine the item ID creation and indexing problem for recommendation foundation models, using P5 as an example of the backbone LLM. To emphasize the importance of item indexing, we first discuss the issues of several trivial item indexing methods, such as random indexing, title indexing, and independent indexing. We then propose four simple yet effective solutions, including sequential indexing, collaborative indexing, semantic (content-based) indexing, and hybrid indexing. Our study highlights the significant influence of item indexing methods on the performance of LLM-based recommendation, and our results on real-world datasets validate the effectiveness of our proposed solutions. The research also demonstrates how recent advances on language modeling and traditional IR principles such as indexing can help each other for better learning and inference. Source code and data are available at https://github.com/Wenyueh/LLM-RecSys-ID.

Short-term load forecasting (STLF) is vital for the effective and economic operation of power grids and energy markets. However, the non-linearity and non-stationarity of electricity demand as well as its dependency on various external factors renders STLF a challenging task. To that end, several deep learning models have been proposed in the literature for STLF, reporting promising results. In order to evaluate the accuracy of said models in day-ahead forecasting settings, in this paper we focus on the national net aggregated STLF of Portugal and conduct a comparative study considering a set of indicative, well-established deep autoregressive models, namely multi-layer perceptrons (MLP), long short-term memory networks (LSTM), neural basis expansion coefficient analysis (N-BEATS), temporal convolutional networks (TCN), and temporal fusion transformers (TFT). Moreover, we identify factors that significantly affect the demand and investigate their impact on the accuracy of each model. Our results suggest that N-BEATS consistently outperforms the rest of the examined models. MLP follows, providing further evidence towards the use of feed-forward networks over relatively more sophisticated architectures. Finally, certain calendar and weather features like the hour of the day and the temperature are identified as key accuracy drivers, providing insights regarding the forecasting approach that should be used per case.

Recent advances in deep learning methods such as LLMs and Diffusion models have created a need for improved quantization methods that can meet the computational demands of these modern architectures while maintaining accuracy. Towards this goal, we study the advantages of FP8 data formats for post-training quantization across 75 unique network architectures covering a wide range of tasks, including machine translation, language modeling, text generation, image classification, generation, and segmentation. We examine three different FP8 representations (E5M2, E4M3, and E3M4) to study the effects of varying degrees of trade-off between dynamic range and precision on model accuracy. Based on our extensive study, we developed a quantization workflow that generalizes across different network architectures. Our empirical results show that FP8 formats outperform INT8 in multiple aspects, including workload coverage (92.64% vs. 65.87%), model accuracy and suitability for a broader range of operations. Furthermore, our findings suggest that E4M3 is better suited for NLP models, whereas E3M4 performs marginally better than E4M3 on computer vision tasks. The code is publicly available on Intel Neural Compressor: https://github.com/intel/neural-compressor.

This paper considers contrastive training for cross-modal 0-shot transfer wherein a pre-trained model in one modality is used for representation learning in another domain using pairwise data. The learnt models in the latter domain can then be used for a diverse set of tasks in a 0-shot way, similar to "Contrastive Language-Image Pre-training (CLIP)" [1] and "Locked-image Tuning (LiT)" [2] that have recently gained considerable attention. Most existing works for cross-modal representation alignment (including CLIP and LiT) use the standard contrastive training objective, which employs sets of positive and negative examples to align similar and repel dissimilar training data samples. However, similarity amongst training examples has a more continuous nature, thus calling for a more'non-binary' treatment. To address this, we propose a novel loss function called Continuously Weighted Contrastive Loss (CWCL) that employs a continuous measure of similarity. With CWCL, we seek to align the embedding space of one modality with another. Owing to the continuous nature of similarity in the proposed loss function, these models outperform existing methods for 0-shot transfer across multiple models, datasets and modalities. Particularly, we consider the modality pairs of image-text and speech-text and our models achieve 5-8% (absolute) improvement over previous state-of-the-art methods in 0-shot image classification and 20-30% (absolute) improvement in 0-shot speech-to-intent classification and keyword classification.

The use of machine learning (ML) models to assess and score textual data has become increasingly pervasive in an array of contexts including natural language processing, information retrieval, search and recommendation, and credibility assessment of online content. A significant disruption at the intersection of ML and text are text-generating large-language models such as generative pre-trained transformers (GPTs). We empirically assess the differences in how ML-based scoring models trained on human content assess the quality of content generated by humans versus GPTs. To do so, we propose an analysis framework that encompasses essay scoring ML-models, human and ML-generated essays, and a statistical model that parsimoniously considers the impact of type of respondent, prompt genre, and the ML model used for assessment model. A rich testbed is utilized that encompasses 18,460 human-generated and GPT-based essays. Results of our benchmark analysis reveal that transformer pretrained language models (PLMs) more accurately score human essay quality as compared to CNN/RNN and feature-based ML methods. Interestingly, we find that the transformer PLMs tend to score GPT-generated text 10-15\% higher on average, relative to human-authored documents. Conversely, traditional deep learning and feature-based ML models score human text considerably higher. Further analysis reveals that although the transformer PLMs are exclusively fine-tuned on human text, they more prominently attend to certain tokens appearing only in GPT-generated text, possibly due to familiarity/overlap in pre-training. Our framework and results have implications for text classification settings where automated scoring of text is likely to be disrupted by generative AI.