As a fundamental problem in transfer learning, model selection aims to rank off-the-shelf pretrained models and select the most suitable one for the new target task. Existing model selection techniques are often constrained in their scope and tend to overlook the nuanced relationships between models and tasks. In this paper, we present a pragmatic framework Fennec, delving into a diverse, large-scale model repository while meticulously considering the intricate connections between tasks and models. The key insight is to map all models and historical tasks into a transfer-related subspace, where the distance between model vectors and task vectors represents the magnitude of transferability. A large vision model, as a proxy, infers a new task's representation in the transfer space, thereby circumventing the computational burden of extensive forward passes and reliance on labels. We also investigate the impact of the inherent inductive bias of models on transfer results and propose a novel method called archi2vec to encode the intricate structures of models. The transfer score is computed through straightforward vector arithmetic with a constant time complexity of O(k) per model inference (where k is the dimension of the latent space). Finally, we make a substantial contribution to the field by releasing a comprehensive benchmark that includes 105 different models. We validate the effectiveness of Fennec through rigorous testing on two benchmarks. Our framework achieves the best evaluation accuracy on both benchmarks. On the PARC benchmark, the feature extraction time is 6.5× faster than the baseline at the same level, and the inference time is 5.7× faster than the fastest baseline. On the larger benchmark, the performance gains are even more pronounced, with feature extraction time being 52.3× faster and inference time 1.8× faster than the fastest baseline, demonstrating remarkable performance efficiency. The code has been made publicly available at: https://github.com/Fay-why/Fennec.

Program performance optimization is essential for achieving high execution efficiency, yet it remains a challenging task that requires expertise in both software and hardware. Large Language Models (LLMs), trained on high-quality code from platforms like GitHub and other open-source sources, have shown promise in generating optimized code for simple snippets. However, current LLM-based solutions often fall short when tackling project-level programs due to the complexity of call graphs and the intricate interactions among functions. In this paper, we emulate the process a human expert might follow when optimizing project-level programs and introduce a three-phase framework POLO (PrOject-Level Optimizer) to address this limitation. First, we profile the program to identify performance bottlenecks using an iterative weighting algorithm. Next, we conduct structural analysis by scanning the project and generating a graph that represents the program's structure. Finally, two LLM agents collaborate in iterative cycles to rewrite and optimize the code at these hotspots, gradually improving performance. We conduct experiments on open-source and proprietary projects. The results demonstrate that POLO accurately identifies performance bottlenecks and successfully applies optimizations. Under the O3 compilation flag, the optimized programs achieved speedups ranging from 1.34x to 21.5x.

The increasing demand for deep neural inference within database environments has driven the emergence of AI?native DBMSs. However, existing solutions either rely on model-centric designs requiring developers to manually select, configure, and maintain models, resulting in high development overhead, or adopt task-centric AutoML approaches with high computational costs and poor DBMS integration. We present MorphingDB, a task-centric AI-native DBMS that automates model storage, selection, and inference within PostgreSQL. To enable flexible, I/O-efficient storage of deep learning models, we first introduce specialized schemas and multi-dimensional tensor data types to support BLOB-based all-in-one and decoupled model storage. Then we design a transfer learning framework for model selection in two phases, which builds a transferability subspace via offline embedding of historical tasks and employs online projection through feature-aware mapping for real-time tasks. To further optimize inference throughput, we propose pre-embedding with vectoring sharing to eliminate redundant computations and DAG-based batch pipelines with cost-aware scheduling to minimize the inference time. Implemented as a PostgreSQL extension with LibTorch, MorphingDB outperforms AI-native DBMSs (EvaDB, Madlib, GaussML) and AutoML platforms (AutoGluon, AutoKeras, AutoSklearn) across nine public datasets, encompassing series, NLP, and image tasks. Our evaluation demonstrates a robust balance among accuracy, resource consumption, and time cost in model selection and significant gains in throughput and resource efficiency.

Obtaining demographic attributes of online users is of great significance for retail marketing, targeted advertisement and many other scenarios. Users' wanderings on various websites and applications contains user preference on different items, and can be leveraged to infer one's private attributes. Existing studies usually focus on manually defined features, relationships in online social networks, or modeling global user preferences. However, attribute inference from the most common behavioral data (e.g., browsing history, shopping cart) is recently overlooked, and still requires further research. In this work, we propose a Topic-Enhanced Attentive Recurrent Neural Network (TEA-RNN) model to capture both local neighborhood-based features (with attentive RNN) and global patterns (with topic model) within user behaviors, and apply multi-task learning mechanism with weighted losses to further leverage the latent relationships within demographics. Experimental results on real-world datasets demonstrates the effectiveness of TEA-RNN by comparing with several commonly used baselines.