conferences

Overcoming Challenges In Automated Image Captioning

At CVPR 2019, IBM researchers introduce an improved method to bridge the semantic gap between visual scenes and language to produce diverse, creative and human-like captions.

Continue reading

Label Set Operations (LaSO) Networks for Multi-Label Few-Shot Learning

Data augmentation is one of the leading methods to tackle the problem of few-shot learning, but current synthesis approaches only address the scenario of a single label per image, when in reality real life images may contain multiple objects. The IBM team came up with a novel technique for synthesizing samples with multiple labels.

Continue reading

RepMet: Representative-Based Metric Learning for Classification and Few-Shot Object Detection

Deep neural networks have demonstrated good results for few-shot learning. However, very few works have investigated the problem of few-shot object detection. A team of IBM researchers developed a novel approach for Distance Metric Learning (DML).

Continue reading

Beyond Backprop: Online Alternating Minimization with Auxiliary Variables

IBM researchers, in collaboration with NYU and MIT, propose a novel alternative to backprop at ICML 2019 that offers competitive performance.

Continue reading

Estimating Information Flow in Deep Neural Networks

Understanding of the macroscopic behavior of deep learning neural networks.

Continue reading

IBM Research AI Moves Machine Learning Forward at ICML 2019

At the 36th International Conference on Machine Learning (ICML 2019), June 10–15 in Long Beach, CA, IBM Research AI will present recent technical advances in machine learning for AI and data science. We’ve led the exploration and development of machine learning technologies for decades, and now we’re progressing the AI field through our portfolio of […]

Continue reading

Exploring the Expressive Range of Conversational Laughter with AI

IBM scientists use crowdsourcing and AI techniques to explore what different types of conversational laughter can tell us.

Continue reading

IBM Sets New Transcription Performance Milestone on Automatic Broadcast News Captioning

IBM sets new performance records for automatic captioning of broadcast news audio, with error rates of 6.5% and 5.9% on two broadcast news benchmarks.

Continue reading

Ultra-Low-Precision Training of Deep Neural Networks

IBM researchers introduce accumulation bit-width scaling, addressing a critical need in ultra-low-precision hardware for training deep neural networks.

Continue reading

Leveraging Temporal Dependency to Combat Audio Adversarial Attacks

A new approach to defend against adversarial attacks in non-image tasks, such as audio input and automatic speech recognition.

Continue reading

Unifying Continual Learning and Meta-Learning with Meta-Experience Replay

Meta-Experience Replay (MER) integrates meta-learning and experience replay to achieve state-of-the-art performance on continual learning benchmarks.

Continue reading

Will Adam Algorithms Work for Me?

A simple and effective approach to monitor the convergence of Adam algorithms, a generic class of adaptive gradient methods for non-convex optimization.

Continue reading