Things to Do in L.A. Tap to enable a layout that focuses on the article. The gift will rename USC's School of Advanced Computing as the USC Mark and Mary Stevens School of Computing and Artificial Intelligence. This is read by an automated voice. Please report any issues or inconsistencies here . USC receives a $200-million gift from venture capitalist Mark Stevens to establish artificial intelligence research and expertise across campus.

We introduce in this paper a novel strategy for efficiently approximating the Sinkhorn distance between twodiscrete measures.

Ensuring the safe exploration of reinforcement learning (RL) agents is critical for deployment in real-world systems. Yet existing approaches struggle to strike the right balance: methods that tightly enforce safety often cripple task performance, while those that prioritize reward leave safety constraints frequently violated, producing diffuse cost landscapes that flatten gradients and stall policy improvement. We introduce the Uncertain Safety Critic (USC), a novel approach that integrates uncertainty-aware modulation and refinement into critic training. By concentrating conservatism in uncertain and costly regions while preserving sharp gradients in safe areas, USC enables policies to achieve effective reward-safety trade-offs. Extensive experiments show that USC reduces safety violations by approximately 40% while maintaining competitive or higher rewards, and reduces the error between predicted and true cost gradients by approximately 83%, breaking the prevailing trade-off between safety and performance and paving the way for scalable safe RL.

The use of audio recordings of human speech to train LLMs poses privacy concerns due to these models' potential to generate outputs that closely resemble artifacts in the training data. In this study, we propose a speaker privacy-preserving representation learning method through the Universal Speech Codec (USC), a computationally efficient encoder-decoder model that disentangles speech into: ( i) privacy-preserving semantically rich representations, capturing content and speech paralinguistics, and ( ii) residual acoustic and speaker representations that enables high-fidelity reconstruction. Extensive evaluations presented show that USC's semantic representation preserves content, prosody, and sentiment, while removing potentially identifiable speaker attributes. Combining both representations, USC achieves state-of-the-art speech reconstruction. Additionally, we introduce an evaluation methodology for measuring privacy-preserving properties, aligning with perceptual tests. We compare USC against other codecs in the literature and demonstrate its effectiveness on privacy-preserving representation learning, illustrating the trade-offs of speaker anonymization, paralinguistics retention and content preservation in the learned semantic representations. Audio samples are shared in https://www.amazon.science/usc-samples . Latest foundational Generative AI (GenAI) revolve around multimodality (Achiam et al., 2023; Anil et al., 2023; Dubey et al., 2024). The extraordinary capabilities of Large Language Models (LLMs) as multimodal learning machines have ushered in a new paradigm for what GenAI can offer to our world (Team, 2025). These foundational LLMs are data-hungry, requiring massive amounts of multimodal training data. Speech and audio are essential modalities for many applications, and mul-timodal models require exposure to them during their training process (Borsos et al., 2023). Speech is a form of individual information (Nautsch et al., 2019), and the development of new foundational speech-aware models demands access to massive amounts of speech data to fully unlock their potential. The research community has collected and curated public data over the past decades, which has been used for specialized speech models (Łajszczak et al., 2024). However, in the realm of Responsible AI, every individual and organization must make proper use of individuals' data when training foundational models, regardless of its public availability. Hence, privacy-preserving methods must be developed to advance foundational speech research while safeguarding individual privacy. Foundational LLMs trained on language modeling tasks model the likelihood of generating coherent text sequences from a distribution of discrete tokens (Touvron et al., 2023). This allows them to produce expressive and varied responses during generation.

Recent work has aimed to improve LLM generations by filtering out hallucinations, thereby improving the precision of the information in responses. Correctness of a long-form response, however, also depends on the recall of multiple pieces of information relevant to the question. In this paper, we introduce Atomic Self-Consistency (ASC), a technique for improving the recall of relevant information in an LLM response. ASC follows recent work, Universal Self-Consistency (USC) in using multiple stochastic samples from an LLM to improve the long-form response. Unlike USC which only focuses on selecting the best single generation, ASC picks authentic subparts from the samples and merges them into a superior composite answer. Through extensive experiments and ablations, we show that merging relevant subparts of multiple samples performs significantly better than picking a single sample. ASC demonstrates significant gains over USC on multiple factoids and open-ended QA datasets - ASQA, QAMPARI, QUEST, ELI5 with ChatGPT and Llama2. Our analysis also reveals untapped potential for enhancing long-form generations using approach of merging multiple samples.

Self-consistency with chain-of-thought prompting (CoT) has demonstrated remarkable performance gains on various challenging tasks, by utilizing multiple reasoning paths sampled from large language models (LLMs). However, self-consistency relies on the answer extraction process to aggregate multiple solutions, which is not applicable to free-form answers. In this work, we propose Universal Self-Consistency (USC), which leverages LLMs themselves to select the most consistent answer among multiple candidates. We evaluate USC on a variety of benchmarks, including mathematical reasoning, code generation, long-context summarization, and open-ended question answering. On open-ended generation tasks where the original self-consistency method is not applicable, USC effectively utilizes multiple samples and improves the performance. For mathematical reasoning, USC matches the standard self-consistency performance without requiring the answer formats to be similar. Finally, without access to execution results, USC also matches the execution-based voting performance on code generation.

A USC sociology, history or dance major may not be attuned to the discipline of quantum computing, but the university's soon to open School of Advanced Computing will open its doors to all -- as well as dramatically expand the number of degrees it confers in technology-related fields, officials announced Thursday. The new University of Southern California school comes at time when jobs for computer and information research scientists are in high demand and fast-growing, projecting to increase 21% from 2021 to 2031, according to the U.S. Bureau of Labor Statistics. A major component of the school will be dedicated to teaching data science and information technology to non-computer majors -- an offering that officials say will allow all students to develop their understanding of elements that shape the digital world, such as coding. "We want to develop a digital backbone across USC that touches every student and every graduate," said Ishwar Puri, senior vice president of research and innovation. "So when they go out into the world, they understand what computing is."

For the past decade in Los Angeles and the State of California, the question is not if there will be wildfires--but rather when and where they will sprout up and how to protect people from these threats. As such, firefighters need to know how to plan and deploy limited resources. One such solution is controlled burns of flammable brush to prevent worst-case scenarios of growing tinder that left unattended, provides fodder for megafires. With $5 million in support from the National Science Foundation's Convergence Accelerator program, a team of researchers, which includes UC San Diego's San Diego Supercomputer Center (SDSC), the University of Southern California's Viterbi School of Engineering and the Tall Timbers Research Station in Florida, will bring the power of AI to help firefighters strategize how best to plan these controlled burns, as well as manage unexpected blazes. SDSC will lead the effort through the development of "BurnPro3D," a new decision support platform to help the fire response and mitigation community quickly and accurately understand risks and tradeoffs presented by a fire to more effectively plan controlled burns and manage wildfires.

USC's Pat Haden and now two "Varsity Blues" defendants want to file briefs in the college admissions scam case under seal. What they want to share, they argue, is "sensitive, confidential, and personally identifiable information." Haden, the former athletic director at the University of Southern California, has filed a motion in federal court in Boston to "quash a trial subpoena for testimony issued by counsel for defendants," as the Herald has reported. He was just granted permission to state his case in private. Defendants Gamal Abdelaziz and John Wilson are seeking that same protection to keep their arguments out of the public eye -- for now.

The University of Southern California and Amazon have joined forces to establish the Center for Secure and Trusted Machine Learning. Housed at USC's Viterbi School of Engineering, the center will provide support to researchers who will focus on developing innovative approaches to privacy-preserving machine learning solutions. This university-industry partnership symbolizes both organizations' shared commitment to "advancing understanding and developing solutions," said USC Provost Charles F. Zukoski in the university's announcement of the partnership. Plans call for the center to leverage talent from both entities in a cooperative effort to unearth new research in this controversial area of study. In USC's announcement, Prem Natarajan, Alexa AI vice president of natural understanding, added, "We are delighted to bring together top talent at Amazon and USC in a joint mission to drive ground-breaking advances in privacy and security preserving machine learning--advances that enable us to continue to safely and securely deliver experiences that enrich and delight our customers worldwide."