The Dreamer algorithm has recently obtained remarkable performance across diverse environment domains by training powerful agents with simulated trajectories. However, the compressed nature of its world model's latent space can result in the loss of crucial information, negatively affecting the agent's performance. Recent approaches, such as $Δ$-IRIS and DIAMOND, address this limitation by training more accurate world models. However, these methods require training agents directly from pixels, which reduces training efficiency and prevents the agent from benefiting from the inner representations learned by the world model. In this work, we propose an alternative approach to world modeling that is both accurate and efficient. We introduce EMERALD (Efficient MaskEd latent tRAnsformer worLD model), a world model using a spatial latent state with MaskGIT predictions to generate accurate trajectories in latent space and improve the agent performance. On the Crafter benchmark, EMERALD achieves new state-of-the-art performance, becoming the first method to surpass human experts performance within 10M environment steps. Our method also succeeds to unlock all 22 Crafter achievements at least once during evaluation.

Mathematical capabilities were previously believed to emerge in common language models only at a very large scale or require extensive math-related pre-training. This paper shows that the LLaMA-2 7B model with common pre-training already exhibits strong mathematical abilities, as evidenced by its impressive accuracy of 97.7% and 72.0% on the GSM8K and MATH benchmarks, respectively, when selecting the best response from 256 random generations. The primary issue with the current base model is the difficulty in consistently eliciting its inherent mathematical capabilities. Notably, the accuracy for the first answer drops to 49.5% and 7.9% on the GSM8K and MATH benchmarks, respectively. We find that simply scaling up the SFT data can significantly enhance the reliability of generating correct answers. However, the potential for extensive scaling is constrained by the scarcity of publicly available math questions. To overcome this limitation, we employ synthetic data, which proves to be nearly as effective as real data and shows no clear saturation when scaled up to approximately one million samples. This straightforward approach achieves an accuracy of 82.6% on GSM8K and 40.6% on MATH using LLaMA-2 7B models, surpassing previous models by 14.2% and 20.8%, respectively. We also provide insights into scaling behaviors across different reasoning complexities and error types.

Transformer, a model comprising attention-based encoder-decoder architecture, have gained prevalence in the field of natural language processing (NLP) and recently influenced the computer vision (CV) space. The similarities between computer vision and medical imaging, reviewed the question among researchers if the impact of transformers on computer vision be translated to medical imaging? In this paper, we attempt to provide a comprehensive and recent review on the application of transformers in medical imaging by; describing the transformer model comparing it with a diversity of convolutional neural networks (CNNs), detailing the transformer based approaches for medical image classification, segmentation, registration and reconstruction with a focus on the image modality, comparing the performance of state-of-the-art transformer architectures to best performing CNNs on standard medical datasets.

It's 1am on the west coast of America, but the Emerald Cloud Lab, just south of San Francisco, is still busy. I'm "visiting" via the camera on a chest-high telepresence robot, being driven round the 1,400 sq metre (15,000 sq ft) lab by Emerald's CEO, Brian Frezza, who is also sitting at home. There are no actual scientists anywhere, just a few staff in blue coats quietly following instructions from screens on their trolleys, ensuring the instruments are loaded with reagents and samples. Cloud labs mean anybody, anywhere can conduct experiments by remote control, using nothing more than their web browser. Experiments are programmed through a subscription-based online interface – software then coordinates robots and automated scientific instruments to perform the experiment and process the data.

Even with the best protocols in place, treating COVID-19 patients is inherently dangerous for health professionals. But what if there was a way to monitor patients from a safe distance? This week a clinical team in Boston has reported being able to monitor a COVID-19 patient remotely, thanks to a device developed at MIT's Computer Science and Artificial Intelligence Laboratory (CSAIL) that can monitor a patient's breathing, movement and sleep patterns using wireless signals. The CSAIL team's device, which they call "Emerald," has been used in multiple hospitals and assistive-care facilities, including with a COVID-19 patient at Heritage Assisted Living in the Boston suburb of Framingham. Developed by MIT professor Dina Katabi and her research group at CSAIL, Emerald is a WiFi-like box that analyzes the wireless signals in the environment using artificial intelligence to infer people's vital signs, sleep, and movement.

Spanish treasure fleets that traversed the Atlantic Ocean to the Americas and back were a 16th-century invention as important as free two-day shipping. Organized 70 years after Columbus's first voyage, the fleet was made up of several specialized ships with one primary goal: Exploiting the riches of the New World as efficiently as possible. The San José, the largest galleon and the flagship of one group of Spanish ships that started sailing in the 16th century, was big and -- thanks to 62 bronze cannons engraved with dolphins -- deadly enough to deter or destroy ships, whether pirates or rival nations. On June 8, 1708, during the War of the Spanish Succession, the San José's gunpowder ignited during a battle with British ships, sending 600 sailors to the bottom of the Atlantic Ocean -- along with gold, silver and emeralds from mines in Peru, a total haul valued at some $17 billion in today's dollars. It stands as one of the most expensive maritime losses in history.

It may not be ready for gift-giving this year, but come 2017 this could be the hottest item on wish lists around the world: a $300 device that enables "X-ray vision." The technology has been under development for more than two years, and now a group of researchers from MIT's Computer Science and Artificial Intelligence Lab (CSAIL) is spinning off a company to market it. Dubbed RF Capture, the technology senses wireless reflections off the human body and can thereby "see" the silhouette of a human standing behind a wall. From across a building, it can determine where you are, who you are, and even which hand you're moving. Through a spin-off called Emerald, the plan is reportedly for it to hit the market in early 2017 at a price between $250 and $300.

Emerald Medical Applications Corp. (OTCQB: MRLA), an Israeli-based company engaged in the development and sale of its proprietary DermaCompare cloud-based, artificial intelligence technology for the early diagnosis of Melanoma/skin cancer, today announced entry into a cooperation agreement with Terem, one of Israel's largest community-based, emergency healthcare providers with 17 medical facilities, serving over 700,000 patients throughout Israel. Starting in April 2016, Emerald will begin to offer its DermaCompare technology at each of Terem's clinics throughout Israel, offering advanced dermatological examinations, diagnosisand treatment led by a leading professional Dermatologists. DermaCompare is Emerald's cloud-based, artificial, intelligence technology using Total Body Photography imaging which is capable of being automatically compared to a patient's previous images to diagnose and detect the presence of Melanoma in its earliest stages. Lior Wayn, Emerald's CEO, stated that "DermaCompare, Emerald's FDA approved, HIPPA compliant software technology, which can be downloaded from any Mac or Android based App store, enables physicians andtheir patients, using virtually any digital camera, including cell phones, iPads, tablets and other similar devices, to take Total Body Photography images and, in real-time, transmit these images for dermatological evaluation and identification of suspicious moles, lesions and other skin conditions. These images are then compared using Emerald's cloud database, as well as the patients previous Total Body Photography images, which will dramatically enhance a physician's ability to detect Melanoma earlier, more accurately and more efficient than other means of diagnosis."

Tech and the elderly are often thought of as being incompatible. Everyone has spent painful hours teaching their grandmother how to text, and entire days have been dedicated to setting up a computer for an elderly person. But tech can actually help the elderly in numerous ways. From smart homes to wireless trackers, tech for the elderly can increase independence, alleviate the concerns of carers and relatives, and generally make life a whole lot easier. The elderly tech market should not be underestimated: the proportion of Britons aged 85 and over is expected to double from 2.5% to 5% within 20 years.

The Semantic Web aims to offer an interoperable environment that will allow users to safely delegate complex actions to intelligent agents. Much work has been done for agents' interoperability; especially in the areas of ontology-based metadata and rule-based reasoning. Nevertheless, the SW proof layer has been neglected so far, although it is vital for agents and humans to understand how a result came about, in order to increase the trust in the interchanged information. This paper focuses on the implementation of third party SW reasoning and proofing services wrapped as agents in a multi-agent framework. This way, agents can exchange and justify their arguments without the need to conform to a common rule paradigm. Via external reasoning and proofing services, the receiving agent can grasp the semantics of the received rule set and check the validity of the inferred results.