Convolutional Neural Networks (CNNs) are widely used for image recognition tasks but are vulnerable to attacks. Most existing attacks create adversarial images of a size equal to the CNN's input size;mainly becaus...
ISBN:
(纸本)9789819958337;9789819958344
Convolutional Neural Networks (CNNs) are widely used for image recognition tasks but are vulnerable to attacks. Most existing attacks create adversarial images of a size equal to the CNN's input size;mainly because creating adversarial images in the high-resolution domain leads to substantial speed, adversity, and visual quality challenges. In a previous work, we developed a method that lifts any existing attack working efficiently in the CNN's input size domain to the high-resolution domain. This method successfully addressed the first two challenges but only partially addressed the third one. The present article provides a crucial refinement of this strategy that, while keeping all its other features, substantially increases the visual quality of the obtained high-resolution adversarial images. The refinement amounts to a blowingup to the high-resolution domain of the adversarial noise created in the low-resolution domain. Adding this blown-up noise to the clean original high-resolution image leads to an almost indistinguishable high-resolution adversarial image. The noise blowing-up strategy is successfully tested on an evolutionary-based black-box targeted attack against VGG-16 trained on ImageNet, with 10 high-resolution clean images.
Automated essay scoring is one of the key applications of natural language processing technology in the field of education. Currently, pre-trained language models for automated essay scoring systems have no significan...
ISBN:
(纸本)9789819947515;9789819947522
Automated essay scoring is one of the key applications of natural language processing technology in the field of education. Currently, pre-trained language models for automated essay scoring systems have no significant advantage over classical models, and pre-trained language models are underutilized for this task. Moreover, in real-world scenarios, supervised models that lack annotated data often perform poorly. To address the issue that the pre-trained language models are not fully applied, we propose a novel prompt tuning model PTAES in this paper, andwe convert the essay scoring procedure into a cloze-style question, after which we design pairs of natural language pattern-verbalizer. Further, we propose a joint model that combines the prompt tuning based model and the pre-trained fine-tuning based model, attempting to address the issue of inadequately supervised models in low-resource situations. Experimental findings indicate that, in supervised, semi-supervised, and zero-shot scenarios, our model can achieve state-of-the-art results, and our method further increases the value of the pre-trained language model in automated essay scoring.
The graph convolutional network has achieved great success since its proposal. Since GCN can be used to study non-Euclidean data, it extends convolutional networks for real-world applications. Graph data is a prevalen...
ISBN:
(数字)9789819947522
ISBN:
(纸本)9789819947515;9789819947522
The graph convolutional network has achieved great success since its proposal. Since GCN can be used to study non-Euclidean data, it extends convolutional networks for real-world applications. Graph data is a prevalent data structure in the real world and is widely used in various fields. Nowadays, most GCN models take data as a complete structure for input. However, real-world data is often incomplete for various reasons, and some data is missing features. Therefore, we propose a GCN model for completing missing data (PGCN) based on the coupled P systems. It can express the missing features of the data using the Gaussian mixture model and attention mechanism. In addition, based on the input, a new activation function is computed in the first layer of the GCN. The proposed PGCN method performs the node classification task on three datasets, and the results show that the method's performance is better than existing missing data processing methods.
Deep learning-based traffic anomaly detection methods are usually fed with high-dimensional statistical features. The greatest challenges are how to detect complex inter-feature relationships and localize and explain ...
ISBN:
(数字)9783031402869
ISBN:
(纸本)9783031402852;9783031402869
Deep learning-based traffic anomaly detection methods are usually fed with high-dimensional statistical features. The greatest challenges are how to detect complex inter-feature relationships and localize and explain anomalies that deviate from these relationships. However, existing methods do not explicitly learn the structure of existing relationships between traffic features or use them to predict the expected behavior of traffic. In this work, we propose a network flow-based IoT anomaly detection approach. It extracts traffic features in different channels as time series. Then a graph neural network combined with a structure learning approach is used to learn relationships between features, which allows users to deduce the root cause of a detected anomaly. We build a real IoT environment and deploy our method on a gateway (simulated with Raspberry PI). The experiment results show that our method has excellent accuracy for detecting anomaly activities and localizes and explains these deviations.
The huge volume of textual information generated in hospitals constitutes an essential but underused asset that could be exploited to improve patient care and management. The encoding of raw medical texts into fixed d...
ISBN:
(纸本)9783031343438;9783031343445
The huge volume of textual information generated in hospitals constitutes an essential but underused asset that could be exploited to improve patient care and management. The encoding of raw medical texts into fixed data structures is traditionally addressed with knowledge-based models and complex hand-crafted rules, but the rigidity of this approach poses limitations to the generalizability and transferability of the solutions, in particular for a non-English setting under data scarcity conditions. This paper shows that transformer-based language representation models have the right characteristics to be employed as a more flexible but equally high-performing clinical information retrieval system for this scenario, without relying upon a knowledge-driven component. We demonstrate it pragmatically on the extraction of clinical entities from Italian cardiology reports for patients with inherited arrhythmias, outperforming the previous ontology-based work with our proposed transformer pipeline under the same setting and exploring a new rule-free approach based on question answering to automate cardiological registry filling.
In classification problems with large output spaces (up to millions of labels), the last layer can require an enormous amount of memory. Using sparse connectivity would drastically reduce the memory requirements, but ...
ISBN:
(纸本)9783031434174;9783031434181
In classification problems with large output spaces (up to millions of labels), the last layer can require an enormous amount of memory. Using sparse connectivity would drastically reduce the memory requirements, but as we show below, applied naively it can result in much diminished predictive performance. Fortunately, we found that this can be mitigated by introducing an intermediate layer of intermediate size. We further demonstrate that one can constrain the connectivity of the sparse layer to be of constant fan-in, in the sense that each output neuron will have the exact same number of incoming connections, which allows for more efficient implementations, especially on GPU hardware. The CUDA implementation of our approach is provided at https://***/xmc-aalto/ecml23-sparse.
In-betweening is the process of drawing transition frames between temporally-sparse keyframes to create a smooth animation sequence. This work presents a novel transformer-based in-betweening technique that serves as ...
ISBN:
(数字)9783031271816
ISBN:
(纸本)9783031271809;9783031271816
In-betweening is the process of drawing transition frames between temporally-sparse keyframes to create a smooth animation sequence. This work presents a novel transformer-based in-betweening technique that serves as a tool for 3D animators. We first show that this problem can be represented as a sequence-to-sequence problem and introduce Tween Transformers - a model that synthesizes high-quality animations using temporally-sparse keyframes as input constraints. We evaluate the model's performance via two complementary methods - quantitative and qualitative evaluation. The model is compared quantitatively with the state-of-the-art models using LaFAN1, a highquality animation dataset. Mean-squared metrics like L2P, L2Q, and NPSS are used for evaluation. Qualitatively, we provide two straightforward methods to assess the model's output. First, we implement a custom ThreeJs-based motion visualizer to render the ground truth, input, and output sequences side by side for comparison. The visualizer renders custom sequences by specifying skeletal positions at temporally-sparse keyframes in JSON format. Second, we build a motion generator to generate custom motion sequences using the model.
We provide a new sequent calculus that enjoys syntactic cut-elimination and strongly terminating backward proof search for the intuitionistic Strong Lob logic iSL, an intuitionistic modal logic with a provability inte...
ISBN:
(纸本)9783031435126;9783031435133
We provide a new sequent calculus that enjoys syntactic cut-elimination and strongly terminating backward proof search for the intuitionistic Strong Lob logic iSL, an intuitionistic modal logic with a provability interpretation. A novel measure on sequents is used to prove both the termination of the naive backward proof search strategy, and the admissibility of cut in a syntactic and direct way, leading to a straight-forward cut-elimination procedure. All proofs have been formalised in the interactive theorem prover Coq.
Intelligent agents are characterized primarily by their farsighted expedient behavior. We present a working prototype of an intelligent agent (ADAM) based on a novel hierarchical neuro-symbolic architecture (Deep Cont...
ISBN:
(纸本)9783031334689;9783031334696
Intelligent agents are characterized primarily by their farsighted expedient behavior. We present a working prototype of an intelligent agent (ADAM) based on a novel hierarchical neuro-symbolic architecture (Deep Control) for deep reinforcement learning with a potentially unlimited planning horizon. The control parameters form a hierarchy of formal languages, where higher-level alphabets contain the semantic meanings of lower-level vocabularies.
Bipartitioning the set of variables Var (Sigma) of a propositional formula Sigma w.r.t. definability consists in pointing out a bipartition of Var (Sigma) such that S defines the variables of O (outputs) in terms of ...
ISBN:
(纸本)9783031436185;9783031436192
Bipartitioning the set of variables Var (Sigma) of a propositional formula Sigma w.r.t. definability consists in pointing out a bipartition < I, O > of Var (Sigma) such that S defines the variables of O (outputs) in terms of the variables in I (inputs), i.e., for every o is an element of O, there exists a formula Phi(o) over I such that o double left right arrow Phi(o) is a logical consequence of Sigma. The existence of Phi(o) given o, I, and Sigma is a coNP-complete problem, and as such, it can be addressed in practice using a SAT solver. From a computational perspective, definability bipartitioning has been shown as a valuable pre-processing technique for model counting, a key task for a number of AI problems involving probabilities. To maximize the benefits offered by such a preprocessing, one is interested in deriving subset-minimal bipartitions in terms of input variables, i.e., definability bipartitions < I, O > such that for every i is an element of I, < I \ {i}, O boolean OR {i}> is not a definability bipartition. We show how the computation of subset-minimal bipartitions can be boosted by leveraging not only the decisions furnished by SAT solvers (as done in previous approaches), but also the SAT witnesses (models and cores) justifying those decisions.
暂无评论