of AI

Selected IBM Research AI publications from 2018

In recent years, machines have met and surpassed human performance in many cognitive tasks. Long-standing challenges in Artificial Intelligence have been conquered.

But are machines truly intelligent? Can AI reach or surpass human capabilities? How can AI augment and scale human expertise and aid us in solving real-world challenges? Much of the recent progress in AI has relied on data-driven techniques like deep learning and artificial neural networks. Given sufficiently large labeled training data sets and enough computation, these approaches are achieving unprecedented results. As a result, there has been a rapid gain on “narrow AI” – tasks in areas such as computer vision, speech recognition, and language translation.

However, a broader set of AI capabilities is needed to progress AI towards solving real-world challenges. In practice, AI systems need to learn effectively and efficiently without large amounts of data. They need to be robust, fair and explainable. They need to integrate knowledge and reasoning together with learning to improve performance and enable more sophisticated capabilities.

Where are we in this evolution?

While "general AI" - AI that can truly think, learn, and reason like a human- is still within the realm of science fiction, "broad AI" that can learn more generally and work across different disciplines is within our reach. IBM Research is driving this evolution. We have been a pioneer of artificial intelligence since the inception of the field, and we continue to expand its frontiers through our portfolio of research focused on three areas: Advancing AI, Scaling AI, and Trusting AI.

Continue on to browse some of our leading research publications from 2018.

Ten noteworthy 2018 publications by the scientists
building the future of AI

Listening Comprehension over Argumentative Content

EMNLP (2018) | Mastering Language

Training Deep Neural Networks with 8-bit Floating Point Numbers

NeurIPS (2018) | Algorithmic Accelerators

Improving Simple
 Models with
 Confidence Profiles

NeurIPS (2018) | Explainability

Delta-Encoder: An Effective Sample Synthesis Method for Few-Shot Object Recognition

NeurIPS (2018) | Augmenting Models

Evidence Aggregation for Answer Re-Ranking in Open-Domain Question Answering

ICLR (2018) | Mastering Language

Evaluating the
Robustness of Neural Networks: An Extreme
 Value Theory Approach

ICLR (2018) | Robustness

Learning to Teach in Cooperative Multiagent Reinforcement Learning

Learning, knowledge, and reasoning

BlockDrop: Dynamic Interference Paths in Residual Networks

CVPR (2018) | Algorithmic Accelerators

Data Pre-Processing
 for Discrimination
 Prevention: Information-
Theoretic Optimization
and Analysis

IEEE Journal of Selected Topics in Signal Processing, August 2018 | Fairness

Deep Learning Architecture Search by Neuro-Cell-based Evolution with Function-Preserving Mutations

ECML-PKDD (2018) | AI Tools and Methodologies