Transfer Learning for Human & AI

An emerging problem in themodern era is fake news detection — many easily available pieces of information are not necessarily factually correct . In this work we explore how differentdocument representations, ranging from simple symbolic bag-of-words, to contextual, neural language model-based ones can be used for efficient fakenews identification .…

Asynchronous iterations are more and more investigated for both scaling and fault-resilience purpose on high performance computing platforms . This paper advocates a novel application direction targeting time-decomposed time-parallel approaches . It turned out that Parareal andasync-Parareal feature very close convergence conditions, asymptoticallyequivalent, including the finite-time termination property .…

SocialVisTUM is a proposed interactive visualization toolkit . It displays a graph with topics asnodes and their correlations as edges . Further details are displayedinteractively to support the exploration of large text collections . The toolkit optimizes automatically on custom data for optimalcoherence.…

Evaluation in NLP is usually done by comparing the scores of competingsystems independently averaged over a common set of test instances . In this work, we question the use of averages for aggregating evaluation scores into afinal number used to decide which system is best .…

GenNI(Generation Negotiation Interface) is an interactive visual system for human-AI collaboration in producing descriptive text . The toolutilizes a deep learning model designed with explicit control states . The visual interface makes it possible for users to interact with AI systems following aRefine-Forecast paradigm to ensure that the generation system acts in a mannerhuman users find suitable .…

How to design a black-box watermarking scheme for automatic speech recognition (ASR) models is still an unsolved problem . The proposed watermarkingscheme is robust against five kinds of attacks and has little impact on accuracy . We synthesize trigger audios by spreading the speechclips of model owners over the entire input audios and labeling the triggeraudios with the stego texts, which hides the authorship information withlinguistic steganography .…

We study coalgebraic modal logic and games to characterise behaviouralequivalence in the presence of side effects . Our aim is to develop a generalframework based on indexed categories/fibrations that is common, at least, to the aforementioned categories . In particular, we show how the coalgebras of behavioural equivalence arises from a relation lifting (a specialkind of indexed morphism) and we give a general recipe to construct suchliftings in the above three cases .…

Shortest path queries over graphs are usually considered as isolated tasks . We define a temporal load-aware network that dynamicallytracks expected loads while satisfying the desirable `first in, first out’property . We develop a scalable collective routing solution that seeks to reduce system-wide congestion through dynamic path reassignment .…

This work studies learning from a synergy process of 3D Morphable Models(3DMM) and 3D facial landmarks to predict complete 3D . facial geometry, including 3D alignment, face orientation, and . 3D landmarks can be extracted and refined from face meshes built by 3DMM parameters .…

A creative image-and-text generative AI system mimics humans’ extraordinary abilities to provide users with diverse and comprehensive caption suggestions . We propose a unified multi-modal framework to achieve this goal . Our framework models image- and-text representations with a Transformer network, which accepts multiple captions as input .…

There has been growing interest in using standard language constructs (e.g. C++’s Parallel Algorithms and Fortran’s do concurrent) foraccelerated computing as an alternative to directive-based APIs . These constructs have the potential to be more portable, and somecompilers already support such standards .…

Fine-grained complexity theory is the area of theoretical computer sciencethat proves conditional lower bounds based on the Strong Exponential TimeHypothesis and similar conjectures . This area has been thriving in the last decade, leading to conditionally best-possible algorithms for a wide variety of problems on graphs, strings, numbers etc.…

This paper considers the noisy group testing problem where among a largepopulation of items some are defective . The goal is to identify all defective items by testing groups of items with the minimum possible number of tests . These algorithms have been used in the decoding of low-density parity-check codes .…

The BEV-SGD improves the robustness of FLOA to Byzantine attacks, by allowing all the workers to send their local updates at the highest transmit power . The theory of Byzantine-resilient FLOA is that it fails to work well with limited defensivecapability to Byzantine attack scenarios .…

The framework of distribution testing is currently ubiquitous in the field of property testing . This work focuses on the connection of the complexities of non-tolerant (“traditional”) testing of distributionsand tolerant testing thereof . When limiting our scope to label-invariant(symmetric) properties of distribution, we prove that the gap is at mostquadratic .…

Each half-duplex (HD)access point (AP) is scheduled to operate in the uplink (UL) or downlink (DL) mode based on the data demands of the user equipments (UEs) The goal is tomaximize the sum UL-DL spectral efficiency (SE) We theoretically establish thesub-modularity of the sum SE, which allows us to develop a new, low complexity,greedy algorithm for the combinatorial AP scheduling problem, with guaranteedoptimality properties .…

The fastest supercomputer in 2020, Fugaku, has achieved digitaltransformation of epidemiology in allowing end-to-end, detailed quantitative quantitative modeling of COVID-19 transmissions for the first time . Fugaku has transformed the behavior of the entire Japanese public through its detailed analysis of transmission risks in multitudes of societal situations entailing heavy risks .…

Given a raw video sequence taken from a freely-moving camera, we study decomposing the observed 3D scene into a static background and adynamic foreground containing the objects that move in the video sequence . We achieve this factorization by reconstructing the video via a triple-stream neuralrendering network that explains the different motions based on corresponding biases .…

Patterns of brain activity are associated with different brain processes and can be used to identify different brain states and make behavioral predictions . However, the relevant features are not readily apparent and accessible . To mineinformative latent representations from multichannel EEG recordings, we propose a novel differentiable EEG decoding pipeline consisting of learnable filters and a pre-determined feature extraction module .…

We study joint learning of image-to-text and text generations, which are naturally bi-directional tasks . We formulate both tasks as sequencegeneration tasks, where we represent images and text as unified sequences oftokens . We further propose two-level granularity feature representations andsequence-level training to improve the Transformer-based unified framework .…

The ambitious CO2 emission targets of the Paris agreements are achievable only with renewable energy, CO2-free power generation, new policies, and planning . The main motivation of this paper is that future green fuels from power-to-X assets should be produced from power with the lowest possible emissions while still keeping the cost of electricity low .…

We study quantum algorithms for several fundamental string problems, including Longest Common Substring, Lexicographically Minimal String Rotation,and Longest Square Substring . These problems have been widely studied in thestringology literature since the 1970s, and are known to be solvable by classical algorithms .…

In recent work, comonads and associated structures have been used to analyze range of important notions in finite model theory, descriptive complexity andcombinatorics . We extend this analysis to Hybrid logic, a widely-studied extension of basic modal logic . In addition to characterising various resource-indexedequivalences induced by Hybrid logic and the bounded fragment, we also give model-theoretic characterisations of bounded formulas in terms of invariance .…

We develop a novel connection between discrepancy minimization and (quantum)communication complexity . We resolve a substantial specialcase of the Matrix Spencer conjecture . We give a polynomial-time algorithm based on partial coloring andsemidefinite programming to find such $x . Our techniques open a new avenue to use tools from communication complexityand information theory to study discrepancy .…

Data heterogeneity across participating devices poses one of the main challenges in federated learning as it has been shown to greatly hamper its convergence time and generalization capabilities . Our approach potentially produces a personalized model for each user at the cost of some extra downlink communication overhead .…

Differentially private algorithms for common metric aggregation tasks often have limited practicality . We propose a simple and practical tool that takes a set of points from an unrestricted (pseudo) metric space as input . The tool can be used to preprocess the input before aggregating it, potentially simplifying the aggregation or boostingits accuracy .…

Social robots need to be able to recognize human needs in real-time so that they can provide timely assistance . We propose anarchitecture that uses social cues to determine when a robot should provide assistance . Enabling a robot to recognize a user’s needs through social cues can help it to adapt to user behaviors andpreferences, which in turn will lead to improved user experiences, say the authors .…

Fine-grained complexity theory is the area of theoretical computer sciencethat proves conditional lower bounds based on the Strong Exponential TimeHypothesis and similar conjectures . This area has been thriving in the last decade, leading to conditionally best-possible algorithms for a wide variety of problems on graphs, strings, numbers etc.…

A paper proposes a sound, abstraction-based technique to establish permutation invariance in DNNs with ReLU as the activation function . The technique computes an over-approximation of the reachable states, and anunder-assessment of the safe states . The novelty of our approach lies in auseful tie-class analysis, that we introduce for forward propagation, and ascalable 2-polytope under-appreciation method that escapes the exponential blow-up in the number of regions during backward propagation .…

This paper proposes two strategies to handle missing data for the classification of electroencephalograms using covariance matrices . The first approach estimates the covariance from imputed data with the $k$-nearestneighbors algorithm . The second relies on the observed data by leveraging theobserved-data likelihood within an expectation-maximization algorithm .…

Inpainting GAN is tasked with learning to reconstruct the original image from the centre crop of images . Images are thenresized rather than cropped and presented as input to the generator . Results show that texture and texture (including canvas and paint) as well as scenery such as clouds, clouds, water, land (including hills and mountains), grass, and flowersare implemented by the generator when extending real artworks .…

In this paper, we propose a patch-based compression process using deep learning, focusing on the lossy point cloudgeometry compression . Unlike existing point cloud compression networks, wedivide the point cloud into patches and compress each patch independently . Our method outperforms the state-of-the-art in terms of rate-distortion performance, especially at lowbitrates .…

We present a convex formulation of compliant frictional contact and a robust,performant method to solve it in practice . Our solver has provenglobal convergence and warm-starts effectively, enabling simulation atinteractive rates . We show our method outperforms existing and open source alternatives without sacrificing accuracy .…

In a PoW-based blockchain network, mining pools compete to successfully mineblocks to pursue rewards . The larger size of a block brings the longerlatency, resulting in a smaller probability of successfully mining . Finding the optimal block size to trade off these two factors is a complex andcrucial problem for the mining pools .…

We study the reward-free RLproblem, where an agent aims to thoroughly explore the environment without any pre-specified reward function . Then, given any extrinsic reward, the agent computes the policy via a planning algorithm with offline data collected in the exploration phase .…

This paper addresses theproblem of cascading errors by focusing on the coupling between the trackingand prediction modules . Rather than relying on a single set of tracking results forprediction, our framework simultaneously reasons about multiple sets oftracking results . We show that this framework improves overall prediction performance over the standard single-hypothesistracking-prediction pipeline by up to 34.2% on the nuScenes dataset, with evenmore significant improvements (up to ~70%) when restricting the evaluation to challenging scenarios involving identity switches and fragments — all with anacceptable computation overhead .…

This paper studies a formalisation of intuitionistic logic by Negri and vonPlato which has general introduction and elimination rules . The philosophical importance of the system is expounded . It is shown that deductions in the systemconvert into normal form and that deductions .…

We propose several heuristics for mitigating one of the main causes of an explosion in rank-based complementation of B\”{u}chi automata(BAs) We introduce two techniques for refiningbounds on the ranks of BA states using data-flow analysis of the automaton . We implement out techniques as an extension of the tool Ranker for BAcomplementation and show that they indeed greatly prune the generated statespace, obtaining significantly better results and outperforming other tools on a large set of benchmarks .…

We show that the fully compressed membership problem for $k$-generated subgroups of $F$ issolvable in polynomial time . We adapt the theory of Stallings’ foldings to handle edges with compressed labels . This partially answers a question of Markus Lohrey .…

We propose a knowledge model for capturing dietary preferences and personalcontext to provide personalized dietary recommendations . We develop a knowledgemodel called the Personal Health Ontology . It represents a patient’s combined medical information, socialdeterminants of health, and observations of daily living elicited frominterviews with diabetic patients .…

Using computational methods, we show that the smallest graph having its eternal domination number less than its cliqueecovering number has $10$ vertices . This answers a question of Klostermeyer andMynhardt [Protecting a graph with mobile guards] In addition, we study the problem ontriangle-free graphs, circulant graphs, planar graphs and cubic graphs .…

Some requests may not be handled properly by the standard natural language understanding (NLU) component . In such cases, a simple reply like “Sorry, I don’t know” hurts the user’s experience . In this paper, we propose a two-stageshortlister-reranker recommender system to match third-party voice applications(skills) to unhandled utterances .…

We develop a novel connection between discrepancy minimization and (quantum)communication complexity . We resolve a substantial specialcase of the Matrix Spencer conjecture . We give a polynomial-time algorithm based on partial coloring andsemidefinite programming to find such $x . Our techniques open a new avenue to use tools from communication complexityand information theory to study discrepancy .…

Each package has an associated reward for delivery and a risk of failure . If the robot fails whiledelivering a package, no future packages can be delivered and the cost of replacing the robot is incurred . We prove anoptimal, greedy solution to both the infinite and finite horizon problems .…

Surface roughness plays an important role in analyzing engineering surfaces . It quantifies the surface topography and can be used to determine whether the resulting surface finish is acceptable or not . We propose a topological Data Analysis (TDA) based approach to classify the roughness level of synthetic surfaces using both their areal images and profiles .…

Alice plans to buy a physical good from Bob over a programmableBlockchain . Alice does not trust Bob, so she is not willing to pay before the good is delivered off-chain . Traditionally, suchscenarios are handled by game-theoretic escrow smart contracts, such asBitHalo .…

Alice plans to buy a physical good from Bob over a programmableBlockchain . Alice does not trust Bob, so she is not willing to pay before the good is delivered off-chain . Traditionally, suchscenarios are handled by game-theoretic escrow smart contracts, such asBitHalo .…

BlockIoT uses blockchain technology to transfer previously inaccessible and centralized data from medical devices to EHR systems . Data tends to be fragmented among healthinfrastructures and prevents interoperability of medical data at the point ofcare . BlockIOT is a suitable system to supplement physicians’ clinical practice and increases efficiency in most healthcare specialties, such as cardiology, pulmonology, endocrinology, and primary care, authors say .…

This paper revisits a classical problem of slotted multiple access with success, idle, and collision events on each slot . An efficient algorithm is developed andconjectured to have an optimal expected capture time for all positive integers $n$. Optimality is proven in the special cases $n \in \{1, 2, 3, 4, 6\}$ using a novel analytical technique that introduces virtual users with enhancedcapabilities .…

The random order streaming model has been very fruitful for graph streams, allowing for polylogarithmic or constant space estimators for fundamentalgraph problems . However, the assumption that there are nocorrelations between the order of arrival of edges in the stream is quitestrong .…