Scientific data spans heterogeneous modalities and is distributed across diverse storage systems. Vector databases have emerged as powerful tools to index high-dimensional embeddings for semantic search, but most existing systems are centralized and focus on indexing a single modality. We introduce SMURF, a system for federated multimodal indexing and retrieval. SMURF aligns pre-computed embeddings from different scientific modalities into a unified embedding space. Our method extends semi-supervised learning techniques to connect disparate embeddings by combining geometry-preserving alignment with weak supervision using a small set of semantically matched data. We evaluate SMURF on cross-modal retrieval tasks using scientific datasets and find that aligned embeddings substantially improve performance—on average, top-5 accuracy nearly doubles (19.18%→38.82%), and nDCG@10, a ranking quality metric, increases by 78% (0.18→0.32) across various retrieval settings. These results highlight the promise of bridging modality gaps to support more effective scientific discovery in federated environments.

Language models (LMs) can “memorize” information, i.e., encode training data in their weights in such a way that inference-time queries can lead to verbatim regurgitation of that data. This ability to extract training data can be problematic, for example, when data are private or sensitive. In this work, we investigate methods to mitigate memorization: three regularizer-based, three fine-tuning-based, and eleven machine unlearning-based methods, with five of the latter being new methods that we introduce. We also introduce TinyMem, a suite of small, computationally-efficient LMs for the rapid development and evaluation of memorization-mitigation methods. We demonstrate that the mitigation methods that we develop using TinyMem can successfully be applied to production-grade LMs, and we determine via experiment that: regularizer-based mitigation methods are slow and ineffective at curbing memorization; fine-tuning-based methods are effective at curbing memorization, but overly expensive, especially for retaining higher accuracies; and unlearning-based methods are faster and more effective, allowing for the precise localization and removal of memorized information from LM weights prior to inference. We show, in particular, that our proposed unlearning method BalancedSubnet outperforms other mitigation methods at removing memorized information while preserving performance on target tasks.

Error-controlled lossy compression has been effective in reducing data storage/transfer costs while preserving reconstructed data fidelity based on user-defined error bounds. State-of-the-art error-controlled lossy compressors primarily fo-cus on error control rather than compression size, and thus, compression ratios are unknown until the compression operation is fully completed. Many use cases, however, require knowledge of compression ratios a priori, for example, pre-allocating appropri-ate memory for the compressed data at runtime. In this paper, we propose a novel, efficient Surrogate-based Error-controlled Lossy Compression Ratio Estimation Framework (SECRE), which includes three key features/contributions. (1) We carefully design the SECRE framework, which, in principle, can be applied to different error-bounded lossy compressors. (2) We implement a compression ratio estimation method for four state-of-the-art error-controlled lossy compressors-SZx, SZ3, ZFP, and SPERR-by devising a corresponding lightweight compression surrogate for each. (3) We evaluate the performance and accuracy of SECRE using four real-world scientific simulation datasets. Experiments show that SECREcan obtain highly accurate com-pression ratio estimates (e.g., ~ 1 % estimation errors for SZx) with low execution overhead (e.g., ~ 2 % estimation cost for SZx).