Jason Lin

Adrià Recasens, Jason Lin, Joāo Carreira et al. · deepmind

Attention-based models are appealing for multimodal processing because inputs from multiple modalities can be concatenated and fed to a single backbone network - thus requiring very little fusion engineering. The resulting representations are however fully entangled throughout the network, which may not always be desirable: in learning, contrastive audio-visual self-supervised learning requires independent audio and visual features to operate, otherwise learning collapses; in inference, evaluation of audio-visual models should be possible on benchmarks having just audio or just video. In this paper, we introduce Zorro, a technique that uses masks to control how inputs from each modality are routed inside Transformers, keeping some parts of the representation modality-pure. We apply this technique to three popular transformer-based architectures (ViT, Swin and HiP) and show that with contrastive pre-training Zorro achieves state-of-the-art results on most relevant benchmarks for multimodal tasks (AudioSet and VGGSound). Furthermore, the resulting models are able to perform unimodal inference on both video and audio benchmarks such as Kinetics-400 or ESC-50.

Stephen Casper, Jason Lin, Joe Kwon et al. · deepmind

Deploying large language models (LMs) can pose hazards from harmful outputs such as toxic or false text. Prior work has introduced automated tools that elicit harmful outputs to identify these risks. While this is a valuable step toward securing models, these approaches rely on a pre-existing way to efficiently classify undesirable outputs. Using a pre-existing classifier does not allow for red-teaming to be tailored to the target model. Furthermore, when failures can be easily classified in advance, red-teaming has limited marginal value because problems can be avoided by simply filtering training data and/or model outputs. Here, we consider red-teaming "from scratch," in which the adversary does not begin with a way to classify failures. Our framework consists of three steps: 1) Exploring the model's range of behaviors in the desired context; 2) Establishing a definition and measurement for undesired behavior (e.g., a classifier trained to reflect human evaluations); and 3) Exploiting the model's flaws using this measure to develop diverse adversarial prompts. We use this approach to red-team GPT-3 to discover classes of inputs that elicit false statements. In doing so, we construct the CommonClaim dataset of 20,000 statements labeled by humans as common-knowledge-true, common knowledge-false, or neither. We are making code and data available.

Yu-Chung Hsiao, Fedir Zubach, Gilles Baechler et al. · deepmind

We introduce ScreenQA, a novel benchmarking dataset designed to advance screen content understanding through question answering. The existing screen datasets are focused either on low-level structural and component understanding, or on a much higher-level composite task such as navigation and task completion for autonomous agents. ScreenQA attempts to bridge this gap. By annotating 86k question-answer pairs over the RICO dataset, we aim to benchmark the screen reading comprehension capacity, thereby laying the foundation for vision-based automation over screenshots. Our annotations encompass full answers, short answer phrases, and corresponding UI contents with bounding boxes, enabling four subtasks to address various application scenarios. We evaluate the dataset's efficacy using both open-weight and proprietary models in zero-shot, fine-tuned, and transfer learning settings. We further demonstrate positive transfer to web applications, highlighting its potential beyond mobile applications.

Maria Wang, Srinivas Sunkara, Gilles Baechler et al.

The rise of powerful multimodal LLMs has enhanced the viability of building web agents which can, with increasing levels of autonomy, assist users to retrieve information and complete tasks on various human-computer interfaces. It is hence necessary to build challenging benchmarks that span a wide-variety of use cases reflecting real-world usage. In this work, we present WebQuest, a multi-page question-answering dataset that requires reasoning across multiple related web pages. In contrast to existing UI benchmarks that focus on multi-step web navigation and task completion, our dataset evaluates information extraction, multimodal retrieval and composition of information from many web pages. WebQuest includes three question categories: single-screen QA, multi-screen QA, and QA based on navigation traces. We evaluate leading proprietary multimodal models like GPT-4V, Gemini Flash, Claude 3, and open source models like InstructBLIP, PaliGemma on our dataset, revealing a significant gap between single-screen and multi-screen reasoning. Finally, we investigate inference time techniques like Chain-of-Thought prompting to improve model capabilities on multi-screen reasoning.

Gheorghe Comanici, Eric Bieber, Mike Schaekermann et al. · amazon-science, baidu

In this report, we introduce the Gemini 2.X model family: Gemini 2.5 Pro and Gemini 2.5 Flash, as well as our earlier Gemini 2.0 Flash and Flash-Lite models. Gemini 2.5 Pro is our most capable model yet, achieving SoTA performance on frontier coding and reasoning benchmarks. In addition to its incredible coding and reasoning skills, Gemini 2.5 Pro is a thinking model that excels at multimodal understanding and it is now able to process up to 3 hours of video content. Its unique combination of long context, multimodal and reasoning capabilities can be combined to unlock new agentic workflows. Gemini 2.5 Flash provides excellent reasoning abilities at a fraction of the compute and latency requirements and Gemini 2.0 Flash and Flash-Lite provide high performance at low latency and cost. Taken together, the Gemini 2.X model generation spans the full Pareto frontier of model capability vs cost, allowing users to explore the boundaries of what is possible with complex agentic problem solving.

Ángel Alexander Cabrera, Fred Hohman, Jason Lin et al.

We present an interactive system enabling users to manipulate images to explore the robustness and sensitivity of deep learning image classifiers. Using modern web technologies to run in-browser inference, users can remove image features using inpainting algorithms and obtain new classifications in real time, which allows them to ask a variety of "what if" questions by experimentally modifying images and seeing how the model reacts. Our system allows users to compare and contrast what image regions humans and machine learning models use for classification, revealing a wide range of surprising results ranging from spectacular failures (e.g., a "water bottle" image becomes a "concert" when removing a person) to impressive resilience (e.g., a "baseball player" image remains correctly classified even without a glove or base). We demonstrate our system at The 2018 Conference on Computer Vision and Pattern Recognition (CVPR) for the audience to try it live. Our system is open-sourced at https://github.com/poloclub/interactive-classification. A video demo is available at https://youtu.be/llub5GcOF6w.

Gilles Baechler, Srinivas Sunkara, Maria Wang et al.

Screen user interfaces (UIs) and infographics, sharing similar visual language and design principles, play important roles in human communication and human-machine interaction. We introduce ScreenAI, a vision-language model that specializes in UI and infographics understanding. Our model improves upon the PaLI architecture with the flexible patching strategy of pix2struct and is trained on a unique mixture of datasets. At the heart of this mixture is a novel screen annotation task in which the model has to identify the type and location of UI elements. We use these text annotations to describe screens to Large Language Models and automatically generate question-answering (QA), UI navigation, and summarization training datasets at scale. We run ablation studies to demonstrate the impact of these design choices. At only 5B parameters, ScreenAI achieves new state-of-the-artresults on UI- and infographics-based tasks (Multi-page DocVQA, WebSRC, MoTIF and Widget Captioning), and new best-in-class performance on others (Chart QA, DocVQA, and InfographicVQA) compared to models of similar size. Finally, we release three new datasets: one focused on the screen annotation task and two others focused on question answering.

Hao-Ping Lee, Yu-Ju Yang, Matthew Bilik et al.

AI creates and exacerbates privacy risks, yet practitioners lack effective resources to identify and mitigate these risks. We present Privy, a tool that guides practitioners through structured privacy impact assessments to: (i) identify relevant risks in novel AI product concepts, and (ii) propose appropriate mitigations. Privy was shaped by a formative study with 11 practitioners, which informed two versions -- one LLM-powered, the other template-based. We evaluated these two versions of Privy through a between-subjects, controlled study with 24 separate practitioners, whose assessments were reviewed by 13 independent privacy experts. Results show that Privy helps practitioners produce privacy assessments that experts deemed high quality: practitioners identified relevant risks and proposed appropriate mitigation strategies. These effects were augmented in the LLM-powered version. Practitioners themselves rated Privy as being useful and usable, and their feedback illustrates how it helps overcome long-standing awareness, motivation, and ability barriers in privacy work.

Chia-Wei Tsai, Chun-Wei Yang, Jason Lin

This study proposes a multiparty mediated quantum secret sharing (MQSS) protocol that allows n restricted quantum users to share a secret via the assistance of a dishonest third-party (TP) with full quantum capabilities. Under the premise that a restricted quantum user can only perform the Hadamard transformation and the Z-basis measurement, the proposed MQSS protocol has addressed two common challenges in the existing semi-quantum secret sharing protocols: (1) the dealer must have full quantum capability, and (2) the classical users must equip with the wavelength quantum filter and the photon number splitters (PNS) to detect the Trojan horse attacks. The security analysis has also delivered proof to show that the proposed MQSS protocol can avoid the collective attack, the collusion attack, and the Trojan horse attacks. In addition, the proposed MQSS protocol is more efficient than the existing SQSS protocols due to the restricted quantum users can only equip with two quantum operations, and the qubits are transmitted within a shorter distance.