In this edition: Can LLMs transform natural language into formal method postconditions; Semantically aligned question + code generation for automated insight generation; Explaining CLIP performance disparities on blind/low vision data; plus recent news.
In this episode, Microsoft’s Chief Questions Officer, Trevor Noah, quizzes Microsoft’s Juan M. Lavista Ferres, Corporate Vice President, Chief Data Scientist, and Director of the AI for Good Lab.
Behind the scenes of maintaining the safety of technology products from harmful and illegal digital content lies unrecognized human labor. The recent rise in the use of generative AI technologies and the accelerating demands to meet responsible AI (RAI) aims necessitates an increased focus on…
Long-term time series forecasting (LTSF) aims to predict future values of a time series given the past values. The current state-of-the-art (SOTA) on this problem is attained in some cases by linear-centric models, which primarily feature a linear mapping layer. However, due to their inherent…
Federated learning enables training high-utility models across several clients without directly sharing their private data. As a downside, the federated setting makes the model vulnerable to various adversarial attacks in the presence of malicious clients. Despite the theoretical and empirical success in defending against attacks…
AI co-creation has the potential to transform the way we interact with virtual worlds and bring about the early VR hopes of speaking shared environments into existence, allowing for novel modes of communicating ideas. The Speaking the world into existence research effort aims to apply the recent…
Suppose we want to train text prediction models in email clients or word processors. The models must preserve the privacy of user data and adhere to a specific fixed size to meet memory and inference time requirements. We introduce a generic framework to solve this…
Secure Two-party Computation (2PC) allows two parties to compute any function on their private inputs without revealing their inputs to each other. In the offline/online model for 2PC, correlated randomness that is independent of all inputs to the computation, is generated in a preprocessing (offline)…
This paper introduces a novel multi-directional speaker, named MuDiS, which utilizes a parametric array to generate highly focused sound beams in multiple directions. The system capitalizes on air nonlinearity to reproduce sound from ultrasounds, successfully overcoming challenges inherent in traditional parametric arrays, such as transducer…
Due to the concise and structured nature of tables, the knowledge contained therein may be incomplete or missing, posing a significant challenge for table question answering (TableQA) systems. However, most existing datasets either overlook the challenge of missing knowledge in TableQA or only utilize unstructured…
Gas sensing plays a key role in detecting explosive/toxic gases and monitoring environmental pollution. Existing approaches usually require expensive hardware or high maintenance cost, and are thus ill-suited for large-scale long-term deployment. In this paper, we propose Gastag, a gas sensing paradigm based on passive…
