ByteDance High-Resolution AMT System Achieves SOTA in Piano Note and Pedal Transcription
ByteDance introduces a high-resolution piano transcription system trained by regressing the precise onset and offset times of piano notes and pedals.
Category: Machine Learning & Data Science
PwC-Powered Code Tab Added to arXiv ML Papers
PwC and arXiv jointly announced their partnership yesterday, unveiling a convenient new Code tab on the abstract page of arXiv Machine Learning articles.
‘Farewell Convolutions’ – ML Community Applauds Anonymous ICLR 2021 Paper That Uses Transformers for Image Recognition at Scale
ICLR 2021 paper An Image Is Worth 16×16 Words: Transformers for Image Recognition at Scale suggests Transformers can outperform top CNNs on CV at scale.
Google, Stanford, & MIT Top NeurIPS 2020 Accepted Papers List
NeurIPS 2020 released its list of accepted papers this week with Google, Stanford, and MIT as the top affiliations.
NVIDIA & Mass General Brigham Hospital Federated Learning Project Predicts COVID-19 Patient Oxygen Need Using 20 Days of Data From 20 Hospitals
NVIDIA, Mass General Brigham and 20 global hospitals launch federated learning initiative EXAM to build AI model for COVID-19 patient oxygen need prediction
Google AI Helps Sign Language ‘Take the Floor’ in Video Conferences
Google AI researchers developed a sign language detection model for video conferencing applications that can perform real-time identification of a person signing as an active speaker.
LPar – A Distributed Multi Agent Platform for Building Polyglot, Omni Channel and Industrial Grade Natural Language Interfaces
LPar, a distributed multi agent platform for large scale industrial deployment of polyglot, diverse and inter-operable agents.
Google AI ‘Looking to Listen’ Tech Enables Speech Enhancement on YouTube Stories in Seconds
Google AI has announced a new audiovisual speech enhancement feature in YouTube Stories (iOS) that enables creators to make better selfie videos by automatically enhancing their voices and reducing noise.
Google, Cambridge, DeepMind & Alan Turing Institute’s ‘Performer’ Transformer Slashes Compute Costs
A team from Google, University of Cambridge, DeepMind, and Alan Turing Institute have proposed a new type of Transformer dubbed Performer, based on a Fast Attention Via positive Orthogonal Random features (FAVOR+) backbone mechanism.
Facebook’s Flexible ‘RAG’ Language Model Achieves SOTA Results on Open-Domain QA
Researchers introduced retrieval-augmented generation - a hybrid, end-to-end differentiable model that combines an information retrieval component with a seq2seq generator.
Google Introduces TensorFlow Recommenders, ‘Helping Users Find What They Love’
Google Brain recently introduced a new open-sourced TensorFlow package, TensorFlow Recommenders designed to simplify the process of building, evaluating, and serving sophisticated recommender models.
Facebook AI Wav2Vec 2.0: Automatic Speech Recognition From 10 Minute Sample
Facebook AI researchers have open-sourced the new wav2vec 2.0 algorithm for self-supervised language learning.
KDD 2020 Celebrates Recipients of the SIGKDD Best Paper Awards
Top Data Scientists Honored for Advanced Research and Applied Data Science in the Field of Knowledge Discovery in Data and Data Mining
Google Approaches BERT-Level Performance Using 300x Fewer Parameters with Extension of Its New NLP model PRADO
The trimmed-down pQRNN extension to Google AI’s projection attention neural network PRADO compares to BERT on text classification tasks for on-device use.
Adobe’s DL-Based ‘HDMatt’ Handles Image Details Thinner Than Hair
UIUC, Adobe Research and University of Oregon propose HDMatt, a Deep Learning-based image matting Cross-Patch Context module for high-resolution image inputs.
UC Berkeley Reward-Free RL Beats SOTA Reward-Based RL
Augmented Temporal Contrast (ATC), a new unsupervised learning (UL) task for learning visual representations agnostic to rewards and without degrading the control policy.
New Google & Oxford Model Time-Shifts People in Videos
A group of researchers from Google Research and the University of Oxford have introduced a novel technique that can “retiming” people’s movements in videos.
After 15 Long Years, a NumPy Paper Finally Appears!
NumPy is the foundation upon which the scientific Python ecosystem is constructed.
New DeepMind Approach ‘Bootstraps’ Self-Supervised Learning of Image Representations
From an augmented view of an image, the researchers trained the online network to predict the target network representation of the same image under a different augmented view.
Czech Tech Does the Monster Mash: Animating 3D Models in Seconds
Monster Mash, a novel AI-powered 3D modelling and animation tool, aims to make these arduous 3D animation processes a whole lot easier.
Facebook AI Enables Automatic Closed Captions for Facebook Live
Facebook AI researchers and engineers just made live video content more accessible by enabling automatic closed captions for Facebook Live and Workplace Live.
Microsoft Democratizes DeepSpeed With Four New Technologies
Microsoft has released four additional DeepSpeed technologies to enable even faster training times, whether on supercomputers or a single GPU.
A Closer Look at the Generalization Gap in Large Batch Training of Neural Networks
This research addresses a well-known phenomenon regarding large batch sizes during training and the generalization gap.
OpenAI ‘GPT-f’ Delivers SOTA Performance in Automated Mathematical Theorem Proving
OpenAI researchers introduce GPT-f, an automated prover and proof assistant for the Metamath formalization language.
TinySpeech: Novel Attention Condensers Enable Deep Recognition Networks on Edge Devices
Novel attention condensers designed to enable the building of low-footprint, highly-efficient deep neural networks for on-device speech recognition on the edge.
New Multitask Benchmark Suggests Even the Best Language Models Don’t Have a Clue What They’re Doing
Researchers introduce a test covering topics such as elementary mathematics, designed to measure language models’ multitask accuracy.
Virginia Tech & Facebook Video Completion Algorithm Achieves SOTA Results
Researchers introduced a novel flow-based video completion algorithm that compares favourably with the state-of-the-art in the field.
Huawei and University of Toronto | All at Once: Temporally Adaptive Multi-Frame Interpolation with Advanced Motion Modeling
This research proposes an efficient and cost-effective solution for multi-frame video interpolation.
A Look at IGLOO: Slicing the Features Space to Represent Sequences
The ReDNA Labs research team has devised a new lightweight approach called IGLOO which allows to deal with sequences up to 25,000 steps long.
Is OpenAI’s GPT-3 API Beta Pricing Too Rich for Researchers?
Although OpenAI hasn’t yet officially announced the GPT-3 pricing scheme, Branwen’s sneak peek has piqued the interest of the NLP community.
DeepMind Uses GNNs to Boost Google Maps ETA Accuracy by up to 50%
DeepMind unveiled a partnership with Google Maps that has leveraged advanced GNNs to improve ETA accuracy.
IIIT Hyperbad’s ‘Wav2Lip’ Boosts Lip-Sync Video Performance
“Wav2Lip,” a novel lip-synchronization model that outperforms current approaches by a large margin in both quantitative metrics and human evaluations.
High-Speed Privacy Protection: Facebook Opacus Trains PyTorch Models With DP
Facebook AI this week released a new high-speed library called Opacus.
AMBERT: BERT with Multi-Grained Tokenization Achieves SOTA Results on English and Chinese NLU Tasks
AMBERT (A Multigrained BERT) leverages both fine-grained and coarse-grained tokenizations to achieve SOTA performance on English and Chinese language tasks.
Deep Claim: Payer Response Prediction from Claims Data with Deep Learning
The paper outlines a predictive model we’ve developed that has the potential to help significantly reduce wasteful healthcare spending.
Can AI Reimagine City Configuration and Automate Urban Planning?
Researchers propose reducing the workloads of urban planners by introducing deep learning systems to handle some of their responsibilities.
Intel Lab Transforms Your Phone into a Robot for $50
Intel Labs researchers have proposed a novel method for building a robot called “OpenBot” on just a US$50 budget.
Detecting Deepfakes: MIT CSAIL Model Identifies Manipulations Using Local Artifacts
Researchers proposed a new model that is designed to spot deepfakes by looking at subtle visual artifacts.
ECCV 2020 Best Paper Award Goes to Princeton Team
The 16th European Conference on Computer Vision (ECCV) kicked off on Sunday as a fully online conference. In the Conference Opening Session this morning, the ECCV organizing committee announced the conference’s paper submission stats and Best Paper selections.
CUHK & Facebook Demonstrate SOTA Monocular 3D Hand Motion Capture Method ‘FrankMocap’
Researchers from The Chinese University of Hong Kong, Facebook Reality Labs, and Facebook AI Research have unveiled a state-of-the-art monocular 3D hand motion capture method, FrankMocap, which can estimate both 3D hand and body motions from in-the-wild monocular inputs with faster speed and better accuracy than previous approaches.