A5027156022- Advanced Bandit Algorithms Research
- Topic Modeling
- Machine Learning and Algorithms
- Natural Language Processing Techniques
- Reinforcement Learning in Robotics
- Speech and dialogue systems
- Optimization and Search Problems
- Human Mobility and Location-Based Analysis
- Smart Grid Energy Management
- Advanced Multi-Objective Optimization Algorithms
- Data Stream Mining Techniques
- Transportation and Mobility Innovations
- Computer Graphics and Visualization Techniques
- Metaheuristic Optimization Algorithms Research
- Age of Information Optimization
- Autonomous Vehicle Technology and Safety
- Traffic Prediction and Management Techniques
- Digital Image Processing Techniques
- Vehicular Ad Hoc Networks (VANETs)
- Auction Theory and Applications
- Urban Transport and Accessibility
- Advanced Vision and Imaging
- 3D Shape Modeling and Analysis
- Transportation Safety and Impact Analysis
- Transportation Planning and Optimization
University of Oxford
2016
University of Southampton
2013-2015
Despite the ubiquity of mobile and wearable text messaging applications, problem keyboard decoding is not tackled sufficiently in light enormous success deep learning Recurrent Neural Network (RNN) Convolutional Networks (CNN) for natural language understanding. In particular, considering that decoders should operate on devices with memory processor resource constraints, makes it challenging to deploy industrial scale neural network (DNN) models. This paper proposes a sequence-to-sequence...
We release the Nemotron-4 340B model family, including Nemotron-4-340B-Base, Nemotron-4-340B-Instruct, and Nemotron-4-340B-Reward. Our models are open access under NVIDIA Open Model License Agreement, a permissive license that allows distribution, modification, use of its outputs. These perform competitively to on wide range evaluation benchmarks, were sized fit single DGX H100 with 8 GPUs when deployed in FP8 precision. believe community can benefit from these various research studies...
As Large Language Models (LLMs) and generative AI become increasingly widespread, concerns about content safety have grown in parallel. Currently, there is a clear lack of high-quality, human-annotated datasets that address the full spectrum LLM-related risks are usable for commercial applications. To bridge this gap, we propose comprehensive adaptable taxonomy categorizing risks, structured into 12 top-level hazard categories with an extension to 9 fine-grained subcategories. This designed...
While large language models (LLMs) have seen unprecedented advancements in capabilities and applications across a variety of use-cases, safety alignment these is still an area active research. The fragile nature LLMs, even that undergone extensive training regimes, warrants additional steering steps via training-free, inference-time methods. recent work the mechanistic interpretability has investigated how activations latent representation spaces may encode concepts, thereafter performed...
Ride sharing has important implications in terms of environmental, social and individual goals by reducing carbon footprints, fostering interactions economizing commuter costs. The ride systems that are commonly available lack adaptive scalable techniques can simultaneously learn from the large scale data predict real-time dynamic fashion. In this paper, we study such a problem towards smart city initiative, where generic system is conceived capable making predictions about share...
We consider a new perspective on dialog state tracking (DST), the task of estimating user's goal through course dialog. By formulating DST as semantic parsing over hierarchical representations, we can incorporate compositionality, cross-domain knowledge sharing and co-reference. present TreeDST, dataset 27k conversations annotated with tree-structured states system acts. describe an encoder-decoder framework for which leads to 20% improvement state-of-the-art approaches that operate flat...
Recent advancements in instruction-tuning datasets have predominantly focused on specific tasks like mathematical or logical reasoning. There has been a notable gap data designed for aligning language models to maintain topic relevance conversations - critical aspect deploying chatbots production. We introduce the CantTalkAboutThis dataset help remain subject at hand during task-oriented interactions. It consists of synthetic dialogues wide range conversation topics from different domains....
As Large Language Models (LLMs) and generative AI become more widespread, the content safety risks associated with their use also increase. We find a notable deficiency in high-quality datasets benchmarks that comprehensively cover wide range of critical areas. To address this, we define broad risk taxonomy, comprising 13 9 sparse categories. Additionally, curate AEGISSAFETYDATASET, new dataset approximately 26, 000 human-LLM interaction instances, complete human annotations adhering to...
Networked data are ubiquitous in this era of the social, economical and technological revolution resting on backbone internet. With spread mobile phones, sensors, embedded devices industrial robots, ability to collect generate interdependent is an all time high. This enormity supplemented with social networks like Facebook, Tumblr, LinkedIn, Twitter many others, that connect people around globe as a network, allowing them share they collect, or distribute, real-time. Further, 'Internet...
Inverse rendering is the problem of decomposing an image into its intrinsic components, i.e. albedo, normal and lighting. To solve this ill-posed from single image, state-of-the-art methods in shape shading mostly resort to supervised training on all components either synthetic or real datasets. Here, we propose a new self-supervised paradigm that 1) reduces need for full supervision decomposition task 2) takes account relighting task. We introduce loss terms leverage consistencies between...
On-line linear optimization on combinatorial action sets (d-dimensional actions) with bandit feedback, is known to have complexity in the order of dimension problem. The exponential weighted strategy achieves best regret bound that $d^{2}\sqrt{n}$ (where $d$ problem, $n$ time horizon). However, such strategies are provably suboptimal or computationally inefficient. attributed structure set and dearth efficient exploration set. Mirror descent entropic regularization function comes close...