Intelligent human inputs are required both in the training and operation of AI systems, and within the governance of blockchain systems and decentralized autonomous organizations (DAOs). This paper presents a formal definition of Human Intelligence Primitives (HIPs), and describes the design and implementation of an Ethereum protocol for their on-chain collection, modeling, and integration in machine learning workflows.
In the last decade, cryptocurrency trading has attracted the attention of private and professional traders and investors. To forecast the financial markets, algorithmic trading systems based on Artificial Intelligence (AI) models are becoming more and more established. However, they suffer from the lack of transparency, thus hindering domain experts from directly monitoring the fundamentals behind market movements. This is particularly critical for cryptocurrency investors, because the study of the main factors influencing cryptocurrency prices, including the characteristics of the blockchain infrastructure, is crucial for driving experts’ decisions. This paper proposes a new visual analytics tool to support domain experts in the explanation of AI-based cryptocurrency trading systems. To describe the rationale behind AI models, it exploits an established method, namely SHapley Additive exPlanations, which allows experts to identify the most discriminating features and provides them with an interactive and easy-to-use graphical interface. The simulations carried out on 21 cryptocurrencies over a 8-year period demonstrate the usability of the proposed tool.
Cryptocurrency blockchain data encounter a class-imbalance problem due to only a few known labels of illicit or fraudulent activities in the blockchain network. For this purpose, we seek to compare various resampling methods applied to two highly imbalanced datasets derived from the blockchain of Bitcoin and Ethereum after further dimensionality reductions, which is different from previous studies on these datasets. Firstly, we study the performance of various classical supervised learning methods to classify illicit transactions or accounts on Bitcoin or Ethereum datasets, respectively. Consequently, we apply various resampling techniques to these datasets using the best performing learning algorithm on each of these datasets. Subsequently, we study the feature importance of the given models, wherein the resampled datasets directly influenced on the explainability of the model. Our main finding is that undersampling using the edited nearest-neighbour technique has attained an accuracy of more than 99% on the given datasets by removing the noisy data points from the whole dataset. Moreover, the best-performing learning algorithms have shown superior performance after feature reduction on these datasets in comparison to their original studies. The matchless contribution lies in discussing the effect of the data resampling on feature importance which is interconnected with explainable artificial intelligence (XAI) techniques.
Blockchain technology has transformed digital transactions into decentralized, transparent, and immutable systems. Smart contracts running on platforms like Ethereum enable automated agreements without intermediaries, but they remain vulnerable to logical errors and security risks that can lead to financial losses. This paper presents a systematic study on AI-based smart contract analysis, using machine learning and deep learning to detect vulnerabilities, anomalies, and potential risks. It proposes a hybrid model combining static analysis (code-level error detection), dynamic analysis (runtime monitoring), and supervised/unsupervised learning techniques. Feature extraction methods convert contract code into formats suitable for AI processing. The approach integrates rule-based systems with AI models to improve detection accuracy and reduce false positives. Evaluation on benchmark datasets shows better performance than traditional methods. The study highlights the effectiveness of AI in enhancing smart contract security and suggests future work on explainable AI, real-time monitoring, and cross-platform interoperability.
Learning to compare two objects are essential in applications, such as digital forensics, face recognition, and brain network analysis, especially when labeled data is scarce and imbalanced. As these applications make high-stake decisions and involve societal values like fairness and transparency, it is critical to explain the learned models. We aim to study post-hoc explanations of Siamese networks (SN) widely used in learning to compare. We characterize the instability of gradient-based explanations due to the additional compared object in SN, in contrast to architectures with a single input instance. We propose an optimization framework that derives global invariance from unlabeled data using self-learning to promote the stability of local explanations tailored for specific query-reference pairs. The optimization problems can be solved using gradient descent-ascent (GDA) for constrained optimization, or SGD for KL-divergence regularized unconstrained optimization, with convergence proofs, especially when the objective functions are nonconvex due to the Siamese architecture. Quantitative results and case studies on tabular and graph data from neuroscience and chemical engineering show that the framework respects the self-learned invariance while robustly optimizing the faithfulness and simplicity of the explanation. We further demonstrate the convergence of GDA experimentally.
Abstract The past few years have witnessed the resurgence of uncertainty estimation generally in neural networks. Providing uncertainty quantification besides the predictive probability is desirable to reflect the degree of belief in the model’s decision about a given input. Recently, Monte-Carlo dropout (MC-dropout) method has been introduced as a probabilistic approach based Bayesian approximation which is computationally efficient than Bayesian neural networks. MC-dropout has revealed promising results on image datasets regarding uncertainty quantification. However, this method has been subjected to criticism regarding the behaviour of MC-dropout and what type of uncertainty it actually captures. For this purpose, we aim to discuss the behaviour of MC-dropout on classification tasks using synthetic and real data. We empirically explain different cases of MC-dropout that reflects the relative merits of this method. Our main finding is that MC-dropout captures datapoints lying on the decision boundary between the opposed classes using synthetic data. On the other hand, we apply MC-dropout method on dataset derived from Bitcoin known as Elliptic data to highlight the outperformance of model with MC-dropout over standard model. A conclusion and possible future directions are proposed.
We introduce SIRUS (Stable and Interpretable RUle Set) for regression, a stable rule learning algorithm which takes the form of a short and simple list of rules. State-of-the-art learning algorithms are often referred to as "black boxes" because of the high number of operations involved in their prediction process. Despite their powerful predictivity, this lack of interpretability may be highly restrictive for applications with critical decisions at stake. On the other hand, algorithms with a simple structure-typically decision trees, rule algorithms, or sparse linear models-are well known for their instability. This undesirable feature makes the conclusions of the data analysis unreliable and turns out to be a strong operational limitation. This motivates the design of SIRUS, which combines a simple structure with a remarkable stable behavior when data is perturbed. The algorithm is based on random forests, the predictive accuracy of which is preserved. We demonstrate the efficiency of the method both empirically (through experiments) and theoretically (with the proof of its asymptotic stability). Our R/C++ software implementation sirus is available from CRAN.
shows that transformers can act as "soft theorem provers" by answering questions over explicitly provided knowledge in natural language. In our work, we take a step closer to emulating formal theorem provers, by proposing PROVER, an interpretable transformer-based model that jointly answers binary questions over rule-bases and generates the corresponding proofs. Our model learns to predict nodes and edges corresponding to proof graphs in an efficient constrained training paradigm. During inference, a valid proof, satisfying a set of global constraints is generated. We conduct experiments on synthetic, hand-authored, and human-paraphrased rule-bases to show promising results for QA and proof generation, with strong generalization performance. First, PROVER generates proofs with an accuracy of 87%, while retaining or improving performance on the QA task, compared to RuleTakers (up to 6% improvement on zero-shot evaluation). Second, when trained on questions requiring lower depths of reasoning, it generalizes significantly better to higher depths (up to 15% improvement). Third, PROVER obtains near perfect QA accuracy of 98% using only 40% of the training data. However, generating proofs for questions requiring higher depths of reasoning becomes challenging, and the accuracy drops to 65% for "depth 5", indicating significant scope for future work.
Machine learning has recently enabled large advances in artificial intelligence, but these results can be highly centralized. The large datasets required are generally proprietary; predictions are often sold on a per-query basis; and published models can quickly become out of date without effort to acquire more data and maintain them. Published proposals to provide models and data for free for certain tasks include Microsoft Research's Decentralized and Collaborative AI on Blockchain. The framework allows participants to collaboratively build a dataset and use smart contracts to share a continuously updated model on a public blockchain. The initial proposal gave an overview of the framework omitting many details of the models used and the incentive mechanisms in real world scenarios. In this work, we evaluate the use of several models and configurations in order to propose best practices when using the Self-Assessment incentive mechanism so that models can remain accurate and well-intended participants that submit correct data have the chance to profit. We have analyzed simulations for each of three models: Perceptron, Naïve Bayes, and a Nearest Centroid Classifier, with three different datasets: predicting a sport with user activity from Endomondo, sentiment analysis on movie reviews from IMDB, and determining if a news article is fake. We compare several factors for each dataset when models are hosted in smart contracts on a public blockchain: their accuracy over time, balances of a good and bad user, and transaction costs (or gas) for deploying, updating, collecting refunds, and collecting rewards. A free and open source implementation for the Ethereum blockchain and simulations written in Python is provided at https://github.com/microsoft/0xDeCA10B. This version has updated gas costs using newer optimizations written after the original publication.
Using blockchain technology, it is possible to create contracts that offer a reward in exchange for a trained machine learning model for a particular data set. This would allow users to train machine learning models for a reward in a trustless manner. The smart contract will use the blockchain to automatically validate the solution, so there would be no debate about whether the solution was correct or not. Users who submit the solutions won't have counterparty risk that they won't get paid for their work. Contracts can be created easily by anyone with a dataset, even programmatically by software agents. This creates a market where parties who are good at solving machine learning problems can directly monetize their skillset, and where any organization or software agent that has a problem to solve with AI can solicit solutions from all over the world. This will incentivize the creation of better machine learning models, and make AI more accessible to companies and software agents.
Artificial intelligence (AI) delivers numerous chances to add to the prosperity of people and the stability of economies and society, yet besides, it adds up a variety of novel moral, legal, social, and innovative difficulties. Trustworthy AI (TAI) bases on the possibility that trust builds the establishment of various societies, economies, and sustainable turn of events, and that people, organizations, and societies can along these lines just at any point understand the maximum capacity of AI, if trust can be set up in its development, deployment, and use. The risks of unintended and negative outcomes related to AI are proportionately high, particularly at scale. Most AI is really artificial narrow intelligence, intended to achieve a specific task on previously curated information from a certain source. Since most AI models expand on correlations, predictions could fail to sum up to various populations or settings and might fuel existing disparities and biases. As the AI industry is amazingly imbalanced, and experts are as of now overpowered by other digital devices, there could be a little capacity to catch blunders. With this article, we aim to present the idea of TAI and its five essential standards (1) usefulness, (2) non-maleficence, (3) autonomy, (4) justice, and (5) logic. We further draw on these five standards to build up a data-driven analysis for TAI and present its application by portraying productive paths for future research, especially as to the distributed ledger technology-based acknowledgment of TAI.
Open access
Artificial Intelligence in Healthcare and Education