The current emissions from computing are almost 4% of the world total. This is already more than emissions from the airline industry and are projected to rise steeply over the next two decades. By 2040 emissions from computing alone will account for more than half of the emission … | Continue reading
Neural Radiance Field training can be accelerated through the use ofgrid-based representations in NeRF's learned mapping from spatial coordinatesto colors and volumetric density. However, these grid-based approaches lack anexplicit understanding of scale and therefore often intro … | Continue reading
We present a systematic study of adversarial attacks on state-of-the-artobject detection frameworks. Using standard detection datasets, we trainpatterns that suppress the objectness scores produced by a range of commonlyused detectors, and ensembles of detectors. Through extensiv … | Continue reading
An expanding civilization could rapidly spread through the galaxy, so theabsence of extraterrestrial settlement in the solar system implies that suchexpansionist civilizations do not exist. This argument, often referred to asthe Fermi paradox, typically assumes that expansion wou … | Continue reading
A Negative Pressure Room (NPR) is an essential requirement by the Bio-SafetyLevels (BSLs) in biolabs or infectious-control hospitals to prevent deadlypathogens from being leaked from the facility. An NPR maintains a negativepressure inside with respect to the outside reference sp … | Continue reading
Machine intelligence is increasingly being linked to claims about sentience,language processing, and an ability to comprehend and transform naturallanguage into a range of stimuli. We systematically analyze the ability ofDALL-E 2 to capture 8 grammatical phenomena pertaining to c … | Continue reading
Many concepts have been proposed for meta learning with neural networks(NNs), e.g., NNs that learn to reprogram fast weights, Hebbian plasticity,learned learning rules, and meta recurrent NNs. Our Variable Shared MetaLearning (VSML) unifies the above and demonstrates that simple … | Continue reading
Video conferencing systems suffer from poor user experience when networkconditions deteriorate because current video codecs simply cannot operate atextremely low bitrates. Recently, several neural alternatives have beenproposed that reconstruct talking head videos at very low bit … | Continue reading
The massive increase of data traffic, the widespread proliferation ofwireless applications and the full-scale deployment of 5G and the IoT, imply asteep increase in cellular networks energy use, resulting in a significantcarbon footprint. This paper presents a comprehensive model … | Continue reading
Recent text-to-image models have achieved impressive results. However, sincethey require large-scale datasets of text-image pairs, it is impractical totrain them on new domains where data is scarce or not labeled. In this work, wepropose using large-scale retrieval methods, in pa … | Continue reading
Proof-of-concept (PoC) of exploits for known vulnerabilities are widelyshared in the security community. They help security analysts to learn fromeach other and they facilitate security assessments and red teaming tasks. Inthe recent years, PoCs have been widely distributed, e.g. … | Continue reading
Modern scientific software stacks have become extremely complex, using manyprogramming models and libraries to exploit a growing variety of GPUs andaccelerators. Package managers can mitigate this complexity using dependencysolvers, but they are reaching their limits. Finding com … | Continue reading
Successful and effective communication between humans and AI relies on ashared experience of the world. By training solely on written text, currentlanguage models (LMs) miss the grounded experience of humans in the real-world-- their failure to relate language to the physical wor … | Continue reading
Neuroscience has long been an important driver of progress in artificialintelligence (AI). We propose that to accelerate progress in AI, we must investin fundamental research in NeuroAI. | Continue reading
Why does time appear to pass irreversibly? To investigate, we introduce aclass of partitioned cellular automata (PCAs) whose cellwise evolution is basedon the chaotic baker's map. After imposing a suitable initial condition andrestricting to a macroscopic view, we are left with a … | Continue reading
Comments | Continue reading
We present UniTune, a simple and novel method for general text-driven imageediting. UniTune gets as input an arbitrary image and a textual editdescription, and carries out the edit while maintaining high semantic andvisual fidelity to the input image. UniTune uses text, an intuit … | Continue reading
A disk-shaped universe (encompassing the observable universe) rotatingglobally with an angular speed equal to the Hubble constant is postulated. Itis shown that dark energy and dark matter are cosmic inertial effects resultingfrom such a cosmic rotation, corresponding to centrifu … | Continue reading
Despite being a key element in the process of disseminating scientificknowledge, editorial notices are often obscured and not clearly linked to thepapers to which they refer. In the present paper, we describe establishedmethods of aggregating notice data, and introduce a novel me … | Continue reading
Recent few-shot methods, such as parameter-efficient fine-tuning (PEFT) andpattern exploiting training (PET), have achieved impressive results inlabel-scarce settings. However, they are difficult to employ since they aresubject to high variability from manually crafted prompts, a … | Continue reading
Cloud applications are increasingly distributing data across multiple regionsand cloud providers. Unfortunately, wide-area bulk data transfers are oftenslow, bottlenecking applications. We demonstrate that it is possible tosignificantly improve inter-region cloud bulk transfer th … | Continue reading
Text-conditioned image editing has recently attracted considerable interest.However, most methods are currently either limited to specific editing types(e.g., object overlay, style transfer), or apply to synthetically generatedimages, or require multiple input images of a common … | Continue reading
Sarcasm Detection has enjoyed great interest from the research community,however the task of predicting sarcasm in a text remains an elusive problem formachines. Past studies mostly make use of twitter datasets collected usinghashtag based supervision but such datasets are noisy … | Continue reading
Trading off exploration and exploitation in an unknown environment is key tomaximising expected return during learning. A Bayes-optimal policy, which doesso optimally, conditions its actions not only on the environment state but onthe agent's uncertainty about the environment. Co … | Continue reading
Context: Security smells are recurring coding patterns that are indicative ofsecurity weakness, and require further inspection. As infrastructure as code(IaC) scripts, such as Ansible and Chef scripts, are used to provisioncloud-based servers and systems at scale, security smells … | Continue reading
Recently, as a generalization of general relativity, a gravity theory hasbeen proposed in which gravitational field equations are described by theCotton tensor. That theory allows an additional contribution to thegravitational potential of a point mass that rises linearly with ra … | Continue reading
The FAIR principles for scientific data (Findable, Accessible, Interoperable,Reusable) are also relevant to other digital objects such as research softwareand scientific workflows that operate on scientific data. The FAIR principlescan be applied to the data being handled by a sc … | Continue reading
RowHammer is a DRAM vulnerability that can cause bit errors in a victim DRAMrow by just accessing its neighboring DRAM rows at a high-enough rate. Recentstudies demonstrate that new DRAM devices are becoming increasingly morevulnerable to RowHammer, and many works demonstrate sys … | Continue reading
Decoding language from brain activity is a long-awaited goal in bothhealthcare and neuroscience. Major milestones have recently been reached thanksto intracranial devices: subject-specific pipelines trained on invasive brainresponses to basic language tasks now start to efficient … | Continue reading
RDF and property graph models have many similarities, such as using basicgraph concepts like nodes and edges. However, such models differ in theirmodeling approach, expressivity, serialization, and the nature of applications.RDF is the de-facto standard model for knowledge graphs … | Continue reading
In this manuscript, we show that any neural network having piece-wise linearactivation functions can be represented as a decision tree. The representationis equivalence and not an approximation, thus keeping the accuracy of theneural network exactly as is. This equivalence shows … | Continue reading
Cloud platforms' rapid growth is raising significant concerns about theircarbon emissions. To reduce emissions, future cloud platforms will need toincrease their reliance on renewable energy sources, such as solar and wind,which have zero emissions but are highly unreliable. Unfo … | Continue reading
I argue against the many-world interpretation (MWI) of quantum theory byemphasizing that when everything is entangled with everything else, in one bigmonstrous piece, there is no room left for creativity. Since the MWI wasinvented, it proves itself wrong (appeared first in French … | Continue reading
Large language models (LLMs) have shown exceptional performance on a varietyof natural language tasks. Yet, their capabilities for HTML understanding --i.e., parsing the raw HTML of a webpage, with applications to automation ofweb-based tasks, crawling, and browser-assisted retri … | Continue reading
This report examines what I see as the core argument for concern aboutexistential risk from misaligned artificial intelligence. I proceed in twostages. First, I lay out a backdrop picture that informs such concern. On thispicture, intelligent agency is an extremely powerful force … | Continue reading
Early backdoor attacks against machine learning set off an arms race inattack and defence development. Defences have since appeared demonstrating someability to detect backdoors in models or even remove them. These defences workby inspecting the training data, the model, or the i … | Continue reading
In response to a recent Nature article which announced an algorithm formultiplying $5\times5$-matrices over $\mathbb{Z}_2$ with only 96multiplications, two fewer than the previous record, we present an algorithmthat does the job with only 95 multiplications. | Continue reading
We introduce GLM-130B, a bilingual (English and Chinese) pre-trained languagemodel with 130 billion parameters. It is an attempt to open-source a 100B-scalemodel at least as good as GPT-3 and unveil how models of such a scale can besuccessfully pre-trained. Over the course of thi … | Continue reading
Data streams are often defined as large amounts of data flowing continuouslyat high speed. Moreover, these data are likely subject to changes in datadistribution, known as concept drift. Given all the reasons mentioned above,learning from streams is often online and under restric … | Continue reading
Internet of Things (IoT) devices are progressively being utilised in avariety of edge applications to monitor and control home and industryinfrastructure. Due to the limited compute and energy resources, activesecurity protections are usually minimal in many IoT devices. This has … | Continue reading
When analyzing large datasets, analysts are often interested in theexplanations for surprising or unexpected results produced by their queries. Inthis work, we focus on aggregate SQL queries that expose correlations in thedata. A major challenge that hinders the interpretation of … | Continue reading
The Transformer is an extremely powerful and prominent deep learningarchitecture. In this work, we challenge the commonly held belief in deeplearning that going deeper is better, and show an alternative design approachthat is building wider attention Transformers. We demonstrate … | Continue reading
We propose a new class of deep reinforcement learning (RL) algorithms thatmodel latent representations in hyperbolic space. Sequential decision-makingrequires reasoning about the possible future consequences of current behavior.Consequently, capturing the relationship between key … | Continue reading
Many developers have ideas to create blockchain applications but do not knowwhere to begin. Often these developers default to using the first blockchaindevelopment platform they discover, which may not be the best platform fortheir project. Over 8000 new blockchain-related projec … | Continue reading
The Move language provides abstractions for programming with digital assetsvia a mix of value semantics and reference semantics. Ensuring memory safety inprograms with references that access a shared, mutable global ledger isdifficult, yet essential for the use-cases targeted by … | Continue reading
Large language models, such as OpenAI's codex and Deepmind's AlphaCode, cangenerate code to solve a variety of problems expressed in natural language.This technology has already been commercialised in at least one widely-usedprogramming editor extension: GitHub Copilot. In this … | Continue reading
The ability to generalise well is one of the primary desiderata of naturallanguage processing NLP). Yet, what `good generalisation' entails and how itshould be evaluated is not well understood, nor are there any common standardsto evaluate it. In this paper, we aim to lay the gro … | Continue reading
Existing permissioned blockchains often rely on coordination-based consensusprotocols to ensure the safe execution of applications in a Byzantineenvironment. Furthermore, these protocols serialize the transactions byordering them into a total global order. The serializability pre … | Continue reading