Morgan Stanley (United States)

Anterior cruciate ligament (ACL) injury remains a high-burden event for adolescent female athletes, with substantial consequences for sport participation, knee health, and long-term quality of life. Exercise-based neuromuscular training (NMT) programs reduce ACL injury rates, but real-world effectiveness is limited by implementation and by the fact that prevention packages are necessarily generic. A persistent open question is whether specific, modifiable physical qualities can be targeted to enhance the mechanical robustness of the knee in the joint configurations and loading rates typical of noncontact injury.This narrative evidence synthesis integrates (i) biomechanical and in vitro evidence on ACL loading mechanisms, (ii) prospective evidence on strength balance and ACL injury risk, and (iii) evidence on hamstring adaptations to eccentric training, including training performed at longer muscle lengths. Based on converging lines of evidence, we propose the working hypothesis that insufficient hamstring eccentric capacity at long muscle lengths (joint angles near knee extension and with hip flexion) is a modifiable contributor to hazardous anterior tibial shear and multiplanar knee loading during rapid deceleration, landing, and cutting. We argue that this capacity is frequently under-assessed and under-trained in youth female athletes.We present a translational prevention framework that specifies (1) candidate field-based assessments, (2) exercise progressions emphasizing long-length hamstring eccentrics integrated with established NMT components, and (3) periodization and implementation considerations for adolescent female athletes. We conclude with a research agenda to test whether adding long-length hamstring eccentric loading to standard NMT improves surrogate biomechanical outcomes and reduces ACL injury incidence.

This article examines two stock-screening methodologies—sequential and simultaneous under a quality investing framework. Using a 21-year sample of US large-cap stocks, this study demonstrates that simultaneous screening yields superior performance, more consistent factor exposures against the benchmark, and lower turnovers in contrast to its sequential counterpart. Through case studies on Microsoft and Apple, the author highlights practical challenges associated with rigid screening thresholds, particularly in benchmark-aware strategies. Importantly, this research underscores structural differences between screening methodologies and their implications for portfolio construction. The findings offer actionable insights for practitioners when selecting or designing screening tools.

Machine Learning (ML) is a relatively recent development that has had a profound impact on Robotics, Computer Vision (CV) and Natural Language Processing (NLP) by allowing automation of intelligent processing, perception and human-computer interaction. ML-driven AI solutions are crucial in these areas for real-time decision-making, pattern recognition, and autonomous operations; traditional rule-based methods tend to lack the adaptability and scalability to meet the demands of these fields. RL and deep learning have increasingly contributed to the improvement of robotic perception, motion planning, and manipulation in robotics, enabling better and more adaptive autonomous systems. Convolutional neural networks (CNNs) and transformer-based architectures have revolutionized computer vision, with applications ranging from object detection and image segmentation to facial recognition, allowing for advanced visual analysis across sectors such as healthcare, security, and self-driving cars. In NLP, the introduction of transformers like BERT and GPT has transformed the space by enabling more context-aware and human-like AI-driven language models, leading to drastic improvements in areas like speech recognition, sentiment analysis, and machine translation. This study delves into the application of ML techniques across these domains, evaluating the implications on efficiency, accuracy, and scalability along with challenges like data quality, model interpretability, and computational constraints. DQN has made advances for robotics, ViT in the domain of computer vision, and LLMs in the subject of NLU. A comparative study to analyse the progress and limitations of such state-of-the-art ML models. The results highlight the promise of multi-modal AI systems, where machine learning algorithms work in conjunction to improve robotic navigation, as well as visual perception and natural language understanding. Extending prior studies of intelligent automation, this research points the way to the development of new adaptive, human-centric, and autonomous AI innovations in diverse sectors.

LLMs have been shown to match or even exceed the performance of specialized Deep Learning models on code generation tasks for general purpose imperative languages, such as Python, Java, C++, and Rust. Conversely, there is only limited work investigating whether such impressive out of the box generalization transfers onto less ubiquitous domain-specific languages, which are often declarative, based on XML, JSON, or YAML. To bridge this gap, we explore the capabilities of LLMs for composing code automation recipes without resorting to any form of task-specific finetuning. We experiment with two GPT versions and CodeLlama-13b-Instruct, and in our experiments, we find that after extensive prompt engineering and chain-of-thought prompting, these models’ accuracy in recipe selection does not go beyond ≈ 30%. For parameter filling of YAML recipes, the accuracy of these models remains below 50%. However, by decomposing the task into two stages: dense retrieval and generative slot filling, and while still keeping our setup training-free, the models are able to attain an accuracy in a range of ≈ 50% to ≈ 67% in recipe selection, and ≈ 60% to ≈ 76% in parameter filling. Our study sheds light on the capabilities of LLMs in generating scripts for less widespread languages and opens up avenues for future research.

This work establishes conditional lower bounds for average-case parity -counting versions of the problems k-XOR, k- SUM, and k-OV. The main contribution is a set of self-reductions for the problems, providing the first specific distributions, for which:

HRMARS - The advent of digital communication tools has fundamentally redefined language education, prompting a need for conceptual exploration of their impact on classroom language practices. This paper investigates the theoretical underpinnings of how digital technologies such as social media, mobile-assisted language learning applications, and Web 2.0 platforms, transform language teaching and learning environments. By synthesising insights from contemporary literature, the paper examines how these tools influence language use, pedagogical approaches, and learners’ engagement. The paper also discusses the conceptual challenges educators face in integrating digital technologies into curriculum, including changes in teachers’ roles and the necessity for new pedagogical skills. This paper aims to provide a comprehensive conceptual framework highlighting the implications of digital communication tools in language education, guiding future research and practice.

While image captioning has gained significant attention, the potential of captioning time-series images, prevalent in areas like finance and healthcare, remains largely untapped. Existing timeseries captioning methods typically offer generic, domain-agnostic descriptions of time-series shapes and struggle to adapt to new domains without substantial retraining. To address these limitations, we introduce TADACap, a retrieval-based framework to generate domain-aware captions for time-series images, capable of adapting to new domains without retraining. Building on TADACap, we propose a novel retrieval strategy that retrieves diverse image-caption pairs from a target domain database, namely TADACap-diverse. We benchmarked TADACap-diverse against state-of-the-art methods and ablation variants. TADACap-diverse demonstrates comparable semantic accuracy while requiring significantly less annotation effort. • Computing methodologies → Visual content-based indexing and retrieval; Natural language generation.

Confusion around the so-called factor zoo is largely due to a failure to distinguish between “attribution” factors and “priced” factors emanating from an asset pricing model. Attribution factors have a zero <italic>expected</italic> mean, do not emanate from asset pricing models, are high in number, can be short term, and should not drive investment policy. Priced factors should have nonzero <italic>expected</italic> premiums, emanate from an asset pricing model, be low in number, be long term, and influence investment policy. Empirical attempts to tame the factor zoo that distinguish between useful, useless, and redundant factors are helpful but could benefit from an overarching theory. The popularity asset pricing model (PAPM), an equilibrium model in which priced factors primarily emanate from the collective tastes of investors, provides a framework for identifying and understanding priced factors, leading to a domesticated factor farm.