There is a rising interest in using artificial intelligence (AI)-powered safety analytics to predict accidents in the trucking industry. Companies may face the practical challenge, however, of not having enough data to develop good safety analytics models. Although pretrained models may offer a solution for such companies, existing safety research using transfer learning has mostly focused on computer vision and natural language processing, rather than accident analytics. To fill the above gap, we propose a pretrain-then-fine-tune transfer learning approach to help any company leverage other companies' data to develop AI models for a more accurate prediction of accident risk. We also develop SafeNet, a deep neural network algorithm for classification tasks suitable for accident prediction. Using the safety climate survey data from seven trucking companies with different data sizes, we show that our proposed approach results in better model performance compared to training the model from scratch using only the target company's data. We also show that for the transfer learning model to be effective, the pretrained model should be developed with larger datasets from diverse sources. The trucking industry may, thus, consider pooling safety analytics data from a wide range of companies to develop pretrained models and share them within the industry for better knowledge and resource transfer. The above contributions point to the promise of advanced safety analytics to make the industry safer and more sustainable.

Secure multiparty computation (MPC) has been proposed to allow multiple mutually distrustful data owners to jointly train machine learning (ML) models on their combined data. However, by design, MPC protocols faithfully compute the training functionality, which the adversarial ML community has shown to leak private information and can be tampered with in poisoning attacks. In this work, we argue that model ensembles, implemented in our framework called SafeNet, are a highly MPC-amenable way to avoid many adversarial ML attacks. The natural partitioning of data amongst owners in MPC training allows this approach to be highly scalable at training time, provide provable protection from poisoning attacks, and provably defense against a number of privacy attacks. We demonstrate SafeNet's efficiency, accuracy, and resilience to poisoning on several machine learning datasets and models trained in end-to-end and transfer learning scenarios. For instance, SafeNet reduces backdoor attack success significantly, while achieving $39\times$ faster training and $36 \times$ less communication than the four-party MPC framework of Dalskov et al. Our experiments show that ensembling retains these benefits even in many non-iid settings. The simplicity, cheap setup, and robustness properties of ensembling make it a strong first choice for training ML models privately in MPC.