Transfer Learning for Human & AI

We prove that Kilian’s four-message succinct argument system is post-quantumsecure in the standard model . At the heart of our proof is a new “measure-and-repair” quantum rewinding procedure that achieves asymptotically optimal knowledge error .…

We address the problem of estimating depth with multi modal audio visual data . Inspired by the ability of animals, such as bats and dolphins, to infer distance of objects with echolocation, some recent methods have utilized echoes for depth estimation .…

Existing state-of-the-art methods for quantifying wear require disassembly and cutting of the examined liner . Adeep-learning framework is proposed that allows computation of the surface-representing bearing load curves from reflection RGB images of the liner surface that can be collected with a simple handheld device .…

We do not know how to write code that is secure in composition with malicious code . We formalize a generaldefinition of reentrancy and introduce a security condition that allows modules like smart contracts to protect their key invariants . We present a security type system that provably enforces secure information flow .…

Message authentication guarantees the integrity of messages exchanged overuntrusted channels . However, the required per-message authentication tagsconsiderably expand packet sizes . We propose R2-D2, which relies on (i)optimal message dependencies, (ii) parametrizable security guarantees, (iii)randomized bit dependencies, and (iv) optional immediate protection bits to address this problem .…

Artificial intelligences are increasingly being embodied and embedded inthe world to carry out tasks and support decision-making with and for people . We define this as the state thatembodied AI “circumstantially” take on within interactive contexts when perceived as both social and agentic by people.…

A Private Set Operation (PSO) protocol involves at least two parties with their private input sets . The goal of the protocol is for the parties to learnt the output of a set operation, i.e. set intersection, on their input sets, without revealing any information about the items that are not in the outputset .…

In large-scale optimization, the presence of nonsmooth and nonconvex terms in a given problem typically makes it hard to solve . Lasry-Lions double envelopes are an extension of the Moreau ones but exhibit an additionalsmoothness property that makes them amenable to fast optimization algorithms .…

Many approaches in the literature perform computationally inefficient full image comparisons between queries and all database images . In this paper, we propose a novel fast sequence-based method for efficient place recognition that can be applied online . It uses relocalization to recover from sequence losses, and exploits usually available but often unused intra-database similarities for potential detection of all matching database images for each query in case of loops or stops in the database .…

Image captioning is a widely known problem in the area of AI . Captiongeneration from floor plan images has applications in indoor path planning,real estate, and providing architectural solutions . The two models take advantage of modern deep neural networks for visual feature extraction and textgeneration .…

Nematode worms are one of most abundant metazoan groups on the earth,occupying diverse ecological niches . Accurate recognition or identification of Nematodes is of great importance for pest control, soil ecology and soil ecology . We propose animage dataset consisting of diverse nematodes (both laboratory cultured andnaturally isolated), which, to our knowledge, is the first time in the community .…

The smartphone and laptop can be unlocked by face or fingerprint recognition, while neural networks which confront numerous requests every day have littlecapability to distinguish between untrustworthy and credible users . In this paper, we propose Model-Lock (M-LOCK) to realize an end-to-endneural network with local dynamic access control, which is similar to the automatic locking function of the smartphone to prevent malicious attackers from obtaining available performance actively when you are away .…

We study text representation methods using deep models . We propose to convert texts into 2-D representations and develop the Sent2Matrix method . Our method allows for the explicit incorporation of both word morphologies and boundaries . This method is the first attempt to represent texts in 2D formats .…

Cybercriminals of today have the tools and the technology to deploy millions of sophisticated attacks . Digital forensics comes into play, but it is not easy to conduct a forensic investigation in IoT systems because of the heterogeneous nature of the IoT environment .…

Multilingual pretrained representations generally rely on subwordsegmentation algorithms to create a shared multilingual vocabulary . Standard heuristic algorithms often lead to sub-optimal segmentation, especially for languages with limited amounts of data . We propose Multi-view Subword Subword Regularization (MVR), a method that enforces the consistencybetween predictions of using inputs tokenized by the standard and probabilisticsegmentations .…

Most existing methods either use RGB-D images, requiring a depth camera, or depending on panoramic photos, assuming that there is little to no occlusion in the rooms . In this work, we proposed GRIHA(Generating Room Interior of a House using ARCore), a framework for generating a layout using an RGB image captured using a simple mobile phone camera .…

Forgery images and videos are usually compressed to different formats such as JPEG and H264 when circulating on the Internet . Existing forgery-detection methods trained on uncompressed data often havesignificantly decreased performance in identifying them . To solve this problem, we propose a novel anti-compression facial forgery detection framework, whichlearns a compression-insensitive embedding feature space utilizing bothoriginal and compressed forgeries .…

This paper provides the first analysis on the feasibility of Return-OrientedProgramming (ROP) on RISC-V . We show the existence of a new class of gadgets, usingseveral Linear Code Sequences And Jumps (LCSAJ) undetected by currentGalileo-based ROP gadget searching tools . We argue that this class is rich enough on Risc-V to mount complex ROP attacks, bypassing traditionalmitigation like DEP, ASLR, stack canaries, G-Free, as well as somecompiler-based backward-edge CFI, by jumping over any guard inserted by acompiler to protect indirect jump instructions .…

Finding the ground state energy of electrons subject to an external electricfield is a fundamental problem in computational chemistry . Schuch and Verstraete haveshown hardness for the electronic-structure problem with an additionalsite-specific external magnetic field, but without the restriction to a fixedbasis .…

A rectilinear Steiner tree for a set $P$ of points in $mathbb{R}^2$ is atree that connects the points using horizontal and vertical linesegments . We present an algorithm with running time $n^{O(\sqrt{\delta) n$ for sparse point sets, that is, point sets where each$1\times\delta$ rectangle inside the strip contains $O(1)$ points .…

The use of local detectors and descriptors in typical computer vision pipelines work well until variations in viewpoint and appearance change becomeextreme . We propose a novel framework that combines learning of invariant descriptorsthrough data augmentation and orthographic viewpoint projection .…

Bayesian neural networks (BNNs) provideuncertainty estimates by averaging predictions with respect to the posteriorweight distribution . Variational inference methods for BNNs approximate theintractable weight posterior with a tractable distribution . They mostly rely on sampling from the variational distribution during training and inference .…

Private distributed learning studies the problem of how multiple distributedentities collaboratively train a shared deep network with their private data . With the security provided by the protocols of blind quantumcomputation, the cooperation between quantum physics and machine learning maylead to unparalleled prospect for solving private distributed learning tasks .…

Deep learning based methods, especially convolutional neural networks (CNNs) have been successfully applied in the field of single image super-resolution(SISR) To obtain better fidelity and visual quality, most of existing networksare of heavy design with massive computation . To this end, this paper explores a novel frequency-awaredynamic network for dividing input into multiple parts according to itscoefficients in the discrete cosine transform (DCT) domain .…

3D-FFS is a novel approach to make sensor fusion based3D object detection networks significantly faster using a class ofcomputationally inexpensive heuristics . Compared to F-ConvNet, we achieve improvements in training andinference times by up to 62.84% and 56.46% . We achieve 0.59%, 2.03% and 3.34%improvements in accuracy for the Car, Pedestrian and Cyclist classes,respectively.…

Cross-lingual self-training (XLST) is able toutilize a small amount of annotated data from high-resource languages to improve representation learning on multilingual un-annotated data . XLST uses a supervised trained model to produce initial representations and another model to learn from them, by maximizing thesimilarity between output embeddings of these two models .…

Blockchain can empower wireless networks with identity management, data integrity, access control, and high-level security . We formalizeBLOWN with the universal composition framework and prove its securityproperties, namely persistence and liveness, against adversarial jamming, double-spending, and Sybil attacks . We propose a novel consensusprotocol named Proof-of-Channel (PoC) leveraging the natural properties ofwireless networks, and a BLOWN protocol (BLOckchain protocol for WirelessNetworks) for wireless networks under an adversarial SINR model .…

Rotationaveraging satisfies strong duality, which enables global solutions to be obtained via semidefinite programming (SDP) relaxation . However, genericsolvers for SDP are rather slow in practice, even on rotation averaging instances of moderate size . In this paper, we present a fast algorithm that achieves global optimality calledrotation coordinate descent (RCD) Unlike block coordinate descent, RCD maintains and updates all valid rotations throughout the iterations .This…

3D object detection plays a crucial role in environmental perception for autonomous vehicles . The Spatial-AttentionGraph Convolution (S-AT GCN) uses the graph convolution and thespatial attention mechanism to extract local geometrical structure features . FE layersboost the pedestrian class performance by 3.62\% and cyclist class by 4.21\% 3DmAP.…

Typical fact verification models use retrieved written evidence to verify claims . Evidence sources often change over time as more information is gathered and revised . In order to adapt, models must be sensitive to subtledifferences in supporting evidence . We present VitaminC, a benchmark infused with challenging cases that require models to discern and adjust to slight factual changes .…

Diabetes is a metabolic disorder that results from defects in autoimmunebeta-cell destruction in Type 1, peripheral resistance to insulin action in Type 2 or, most commonly, both . Patients with long-standing diabetes often fall prey to Diabetic Retinopathy (DR) resulting in changes in the retina of the human eye, which may lead to loss of vision in extreme cases .…

The UrbanVCAsimulates multiple types of urban land-use changes at the land-parcel level have achieved a high accuracy (FoM=0.243) and the landscape index similarityreaches 87.3% . The simulation results in 2030 show that the eco-protectionscenario can promote urban agglomeration and reduce ecological aggression and loss of arable land by at least 60% .…

We study families of depth measures defined by natural sets of axioms . Weshow that any such depth measure is a constant factor approximation of Tukeydepth . We further investigate the dimensions of depth regions, showing that theCascade conjecture, introduced by Kalai for Tverberg depth, holds for all depthmeasures .…

BGPeek-a-Boo monitors DDoS attacks with honeypots and uses BGP poisoning to temporarily shutdown ingress traffic from selected Autonomous Systems . We then show how a graph-based model of BGP route propagation can reduce the search space, resulting in a 5x median speed-up and over 20x for 1/4 of all cases .…

Intrusion Detection Systems (IDS) are now an essential element when it comesto securing computers and networks . Despite the huge research efforts done inthe field, handling sources’ reliability remains an open issue . To address thisproblem, this paper proposes a novel contextual discounting method based on reliability and their distinguishing ability between normal andnormal behavior .…

A novel marginpreserving self-paced contrastive Learning (MPSCL) model for cross-modal medical image segmentation . MPSCL significantlyimproves semantic segmentation performance, and outperforms a wide variety of state-of-the-art methods by a large margin . Unlike the conventional construction of contrastivepairs in contrastive learning, the domain-adaptive category prototypes are used to constitute the positive and negative sample pairs .…

Payment Protocol standard BIP70, specifying how payments in Bitcoin are performed by merchants and customers, is supported by the largest paymentprocessors and most widely-used wallets . The Payment Protocol has been shown to be vulnerable to refund attacks due to lack of authentication of refund addresses .…

This work presents a novel approach to efficiently model anodic dissolution in electrochemical machining . Earlier modeling approaches employ a strict spacediscretization of the anodic surface that is associated with a remeshingprocedure at every time step . This inner variable allows the modeling of the complex dissolution process without the necessity of computationally expensiveremeshing by controlling the effective material parameters .…

We connect the study of pseudodeterministic algorithms to two major open problems about the structural complexity of $BPTIME . We build on techniques developed to prove hierarchy theorems for probabilistic time with advice (Fortnow and Santhanam, FOCS 2004) to construct the first unconditional pseudorandom generator of polynomial stretch computable .…

We investigate the benefits of feature selection, nonlinear modelling and online learning with forecasting in financial time series . We also find that, in the subset of models we use,sequential learning in time with online Ridge regression, provides the best multi-step ahead forecasts, and continual learning with an online radial basisfunction network .…

Inferring the probability distribution of sentences or word sequences is akey process in natural language processing . The proposed NCE samples negative sentences independently of aprevious text so that the trained model gives higher probabilities to thesentences that are more consistent with the context .…

Given a finite set $A\subset\mathbb{R}^d, let Cov$_{_{r,k}$ denote the set ofall points within distance $r$ to at least $k$ points of $A$. Allowing $r and$k$ to vary, we obtain a 2-parameter family of spaces that grow larger when $r# increases or $k# decreases .…

Several attacks have been proposed against Proof-of-Work blockchains . decreasing the profitability of mining forhonest nodes incentivizes them to stop mining or to leave the attacked chain . The departure of honest nodes exacerbates the attack and may further decrease profitability and incentivize more honest nodes to leave .…

In this paper, we thoroughly study therobustness of uncompressed, distilled, pruned and binarized neural networks against white-box and black-box adversarial attacks . These new insights facilitate defensive trainingschemes or reactive filtering methods, where the attack is detected and the input is discarded and/or cleaned .…

For a set $S$ of $n$ disjoint line segments in $\mathbb{R}^{2}$, thevisibility counting problem is to preprocess $S $S such that the number ofvisible segments in $S# can be computed quickly . Therehave been approximation algorithms for this problem with trade off betweenspace and query time .…

The NLP field is becoming increasingly interested in, and conscientious about, the reproducibility of its results . The past few years have seen an impressive range of new initiatives, events and active research inthe area . However, the field is far from reaching a consensus about howreproducibility should be defined, measured and addressed, with diversity of views currently increasing rather than converging .…

Vision-language pre-training (VLP) on large-scale image-text pairs has witnessed rapid progress for learning cross-modal representations . SemVLP jointly aligns both the low-level and high-level semantics between image and text representations . An extensive set of experiments have been conducted on four well-established vision-language understanding tasks todemonstrate the effectiveness of the proposed model .…