Inference accuracy of deep neural networks (DNNs) is a crucial performance metric, but can vary greatly in practice subject to actual test datasets and is typically unknown due to the lack of ground truth labels. This has raised significant concerns
with trustworthiness of DNNs, especially in safety-critical applications. In this paper, we address trustworthiness of DNNs by using post-hoc processing to monitor the true inference accuracy on a users dataset. Concretely, we propose a neural network-based accuracy monitor model, which only takes the deployed DNNs softmax probability output as its input and directly predicts if the DNNs prediction result is correct or not, thus leading to an estimate of the true inference accuracy. The accuracy monitor model can be pre-trained on a dataset relevant to the target application of interest, and only needs to actively label a small portion (1% in our experiments) of the users dataset for model transfer. For estimation robustness, we further employ an ensemble of monitor models based on the Monte-Carlo dropout method. We evaluate our approach on different deployed DNN models for image classification and traffic sign detection over multiple datasets (including adversarial samples). The result shows that our accuracy monitor model provides a close-to-true accuracy estimation and outperforms the existing baseline methods.
To increase the trustworthiness of deep neural network (DNN) classifiers, an accurate prediction confidence that represents the true likelihood of correctness is crucial. Towards this end, many post-hoc calibration methods have been proposed to lever
age a lightweight model to map the target DNNs output layer into a calibrated confidence. Nonetheless, on an out-of-distribution (OOD) dataset in practice, the target DNN can often mis-classify samples with a high confidence, creating significant challenges for the existing calibration methods to produce an accurate confidence. In this paper, we propose a new post-hoc confidence calibration method, called CCAC (Confidence Calibration with an Auxiliary Class), for DNN classifiers on OOD datasets. The key novelty of CCAC is an auxiliary class in the calibration model which separates mis-classified samples from correctly classified ones, thus effectively mitigating the target DNNs being confidently wrong. We also propose a simplified version of CCAC to reduce free parameters and facilitate transfer to a new unseen dataset. Our experiments on different DNN models, datasets and applications show that CCAC can consistently outperform the prior post-hoc calibration methods.
Attacks based on power analysis have been long existing and studied, with some recent works focused on data exfiltration from victim systems without using conventional communications (e.g., WiFi). Nonetheless, prior works typically rely on intrusive
direct power measurement, either by implanting meters in the power outlet or tapping into the power cable, thus jeopardizing the stealthiness of attacks. In this paper, we propose NoDE (Noise for Data Exfiltration), a new system for stealthy data exfiltration from enterprise desktop computers. Specifically, NoDE achieves data exfiltration over a buildings power network by exploiting high-frequency voltage ripples (i.e., switching noises) generated by power factor correction circuits built into todays computers. Located at a distance and even from a different room, the receiver can non-intrusively measure the voltage of a power outlet to capture the high-frequency switching noises for online information decoding without supervised training/learning. To evaluate NoDE, we run experiments on seven different computers from top-vendors and using top brand power supply units. Our results show that for a single transmitter, NoDE achieves a rate of up to 28.48 bits/second with a distance of 90 feet (27.4 meters) without the line of sight, demonstrating a practically stealthy threat. Based on the orthogonality of switching noise frequencies of different computers, we also demonstrate simultaneous data exfiltration from four computers using only one receiver. Finally, we present a few possible defenses, such as installing noise filters, and discuss their limitations.
Two-dimensional carbides and nitrides of transition metals, known as MXenes, are a fast-growing family of 2D materials that draw attention as energy storage materials. So far, MXenes are mainly prepared from Al-containing MAX phases (where A = Al) by
Al dissolution in F-containing solution, but most other MAX phases have not been explored. Here, a redox-controlled A-site-etching of MAX phases in Lewis acidic melts is proposed and validated by the synthesis of various MXenes from unconventional MAX phase precursors with A elements Si, Zn, and Ga. A negative electrode of Ti3C2 MXene material obtained through this molten salt synthesis method delivers a Li+ storage capacity up to 738 C g-1 (205 mAh g-1) with high-rate performance and pseudocapacitive-like electrochemical signature in 1M LiPF6 carbonate-based electrolyte. MXene prepared from this molten salt synthesis route offer opportunities as high-rate negative electrode material for electrochemical energy storage applications.
We study effects of disorder (randomness) in a 2D square-lattice $S=1/2$ quantum spin system, the $J$-$Q$ model with a 6-spin interaction $Q$ supplementing the Heisenberg exchange $J$. In the absence of disorder the system hosts antiferromagnetic (AF
M) and columnar valence-bond-solid (VBS) ground states. The VBS breaks $Z_4$ symmetry, and in the presence of arbitrarily weak disorder it forms domains. Using QMC simulations, we demonstrate two kinds of such disordered VBS states. Upon dilution, a removed site leaves a localized spin in the opposite sublattice. These spins form AFM order. For random interactions, we find a different state, with no order but algebraically decaying mean correlations. We identify localized spinons at the nexus of domain walls between different VBS patterns. These spinons form correlated groups with the same number of spinons and antispinons. Within such a group, there is a strong tendency to singlet formation, because of spinon-spinon interactions mediated by the domain walls. Thus, no long-range AFM order forms. We propose that this state is a 2D analog of the well-known 1D random singlet (RS) state, though the dynamic exponent $z$ in 2D is finite. By studying the T-dependent magnetic susceptibility, we find that $z$ varies, from $z=2$ at the AFM--RS phase boundary and larger in the RS phase The RS state discovered here in a system without geometric frustration should correspond to the same fixed point as the RS state recently proposed for frustrated systems, and the ability to study it without Monte Carlo sign problems opens up opportunities for further detailed characterization of its static and dynamic properties. We also discuss experimental evidence of the RS phase in the quasi-two-dimensional square-lattice random-exchange quantum magnets Sr$_2$CuTe$_{1-x}$W$_x$O$_6$.
We study the Neel-paramagnetic quantum phase transition in two-dimensional dimerized $S=1/2$ Heisenberg antiferromagnets using finite-size scaling of quantum Monte Carlo data. We resolve the long standing issue of the role of cubic interactions arisi
ng in the bond-operator representation when the dimer pattern lacks a certain symmetry. We find non-monotonic (monotonic) size dependence in the staggered (columnar) dimerized model, where cubic interactions are (are not) present. We conclude that there is an irrelevant field in the staggered model that is not present in the columnar case, but, at variance with previous claims, it is not the leading irrelevant field. The new exponent is $omega_2 approx 1.25$ and the prefactor of the correction $L^{-omega_2}$ is large and comes with a different sign from that of the formally leading conventional correction with exponent $omega_1 approx 0.78$. Our study highlights the possibility of competing scaling corrections at quantum critical points.
The theory of deconfined quantum critical points describes phase transitions at temperature T = 0 outside the standard paradigm, predicting continuous transformations between certain ordered states where conventional theory requires discontinuities.
Numerous computer simulations have offered no proof of such transitions, however, instead finding deviations from expected scaling relations that were neither predicted by the DQC theory nor conform to standard scenarios. Here we show that this enigma can be resolved by introducing a critical scaling form with two divergent length scales. Simulations of a quantum magnet with antiferromagnetic and dimerized ground states confirm the form, proving a continuous transition with deconfined excitations and also explaining anomalous scaling at T > 0. Our findings revise prevailing paradigms for quantum criticality, with potentially far-reaching implications for many strongly-correlated materials.
We study the mechanism of decay of a topological (winding-number) excitation due to finite-size effects in a two-dimensional valence-bond solid state, realized in an $S=1/2$ spin model ($J$-$Q$ model) and studied using projector Monte Carlo simulatio
ns in the valence bond basis. A topological excitation with winding number $|W|>0$ contains domain walls, which are unstable due to the emergence of long valence bonds in the wave function, unlike in effective descriptions with the quantum dimer model. We find that the life time of the winding number in imaginary time diverges as a power of the system length $L$. The energy can be computed within this time (i.e., it converges toward a quasi-eigenvalue before the winding number decays) and agrees for large $L$ with the domain-wall energy computed in an open lattice with boundary modifications enforcing a domain wall. Constructing a simplified two-state model and using the imaginary-time behavior from the simulations as input, we find that the real-time decay rate out of the initial winding sector is exponentially small in $L$. Thus, the winding number rapidly becomes a well-defined conserved quantum number for large systems, supporting the conclusions reached by computing the energy quasi-eigenvalues. Including Heisenberg exchange interactions which brings the system to a quantum-critical point separating the valence-bond solid from an antiferromagnetic ground state (the putative deconfined quantum-critical point), we can also converge the domain wall energy here and find that it decays as a power-law of the system size. Thus, the winding number is an emergent quantum number also at the critical point, with all winding number sectors becoming degenerate in the thermodynamic limit. This supports the description of the critical point in terms of a U(1) gauge-field theory.
A graph $G$ is emph{uniquely k-colorable} if the chromatic number of $G$ is $k$ and $G$ has only one $k$-coloring up to permutation of the colors. A uniquely $k$-colorable graph $G$ is edge-critical if $G-e$ is not a uniquely $k$-colorable graph for
any edge $ein E(G)$. Melnikov and Steinberg [L. S. Melnikov, R. Steinberg, One counterexample for two conjectures on three coloring, Discrete Math. 20 (1977) 203-206] asked to find an exact upper bound for the number of edges in a edge-critical 3-colorable planar graph with $n$ vertices. In this paper, we give some properties of edge-critical uniquely 3-colorable planar graphs and prove that if $G$ is such a graph with $n(geq6)$ vertices, then $|E(G)|leq frac{5}{2}n-6 $, which improves the upper bound $frac{8}{3}n-frac{17}{3}$ given by Matsumoto [N. Matsumoto, The size of edge-critical uniquely 3-colorable planar graphs, Electron. J. Combin. 20 (3) (2013) $#$P49]. Furthermore, we find some edge-critical 3-colorable planar graphs which have $n(=10,12, 14)$ vertices and $frac{5}{2}n-7$ edges.
