Therefore, toensure that both the expected per-iteration cost and the variance are finite, itrequires thatc

PL 1 i=l fRT(xi)) would be, and another term of E XL( Xl PL 1 i=l fRT(Xi)) that propagates through theTransformer blocks. Figure 1 shows the full comparison of the baseline and PLD, fine-tuned at different checkpoints. Overall, we observe that PLD not only trains BERT faster in pre-training but also preserves the performanceondownstreamtasks. Results are visualized in Figure 1, which shows that the baseline is less robust on the choice of learningrates.

F s,iBs,i, n = 1,2,...,S, (10) where F s is the support features extracted by a pretrained ViT. Inspired by the multiple-object tracking within a single framework [21], in which different objects are represented by various identifications (i.e., learnable vectors) for simultaneously tracking, we add extra learnable tokens tothemeanfeatures formorediscriminativeprompts.

We note that eachxli,t influences Et in two ways: (i) it occurs in Eq.(6) explicitly, but (ii) it also determinesthevaluesof µl 1k,t viaEq.(1). For the experiments on MHNs, the parameterβ was extensively tested, as we have usedβ {1,2,3,5,10,100,1000},and always reported the best result. For experiments comparing against classical Hopfield networks, we have convertedeveryimagetobinary. Results: The results, plotted inFigure 1, are similar tothe ones ofCIFAR10. So far, we analyzed images with Gaussian noise of variance0.2.

Self-supervised learning in computer vision trains on unlabeled data, such as images or (image, text) pairs, to obtain an image encoder that learns high-quality embeddings for input data. Emerging backdoor attacks towards encoders expose crucial vulnerabilities of self-supervised learning, since downstream classifiers (even further trained on clean data) may inherit backdoor behaviors from encoders. Existing backdoor detection methods mainly focus on supervised learning settings and cannot handle pre-trained encoders especially when input labels are not available. In this paper, we propose DECREE, the first backdoor detection approach for pre-trained encoders, requiring neither classifier headers nor input labels. We evaluate DECREE on over 400 encoders trojaned under 3 paradigms. We show the effectiveness of our method on image encoders pre-trained on ImageNet and OpenAI's CLIP 400 million image-text pairs. Our method consistently has a high detection accuracy even if we have only limited or no access to the pre-training dataset.

TECHNICAL NOTE - TN-2023-02-28, FEBRUARY 2023. 1 Abstract This short technical note points out an erroneous claim about a new rule of combination of basic belief assignments presented recently by Kenn et al. in [1], referred as Kenn's rule of combination (or just as KRC for short). We prove thanks a very simple counter-example that Kenn's rule is not associative. Consequently, the results of the method proposed by Kenn et al. highly depends on the ad-hoc sequential order chosen for the fusion process as proposed by the authors. This serious problem casts in doubt the interest of this method and its real ability to provide trustful results and to make good decisions to help for precise breast cancer therapy. Recently a paper devoted to the Breast Cancer Precision Therapy by Kenn et al. [1] attracted our attention for two main reasons: 1) this application of information fusion is very interesting and important; 2) Kenn's et al. method is based on a new rule of combination of basic belief assignments (BBAs).