/*! elementor – v3.6.6 – 08-06-2022 */
.elementor-widget-image{text-align:center}.elementor-widget-image a{display:inline-block}.elementor-widget-image a img[src$=”.svg”]{width:48px}.elementor-widget-image img{vertical-align:middle;display:inline-block} 
/*! elementor – v3.6.6 – 08-06-2022 */
.e-container.e-container–row .elementor-spacer-inner{width:var(–spacer-size)}.e-container.e-container–column .elementor-spacer-inner,.elementor-column .elementor-spacer-inner{height:var(–spacer-size)}
- New machine learning algorithm finds a gene signature characteristic of tumors
- ‘VALHALLA’, a Machine Learning Method That can Hallucinate an Image of Written Words, and Then Use It to Help Translate The Text into Another Language
- How AI Is Useful — and Not Useful — for Cybersecurity
- How AI is shaping the future of work
New machine learning algorithm finds a gene signature characteristic of tumors
https://www.news-medical.net/news/20220610/New-machine-learning-algorithm-finds-a-gene-signature-characteristic-of-tumors.aspx
- Emily Henderson
- June 10, 2022
Notes:
A new machine learning algorithm called “ikarus” has found a gene signature characteristic of tumors. To reliably distinguish cancer cells from healthy cells, a team led by Dr. Altuna Akalin, head of the Bioinformatics and Omics Data Science Platform at the Max Delbrück Center for Molecular Medicine in the Helmholtz Association (MDC), has now developed a machine learning program called “ikarus.” The program found a pattern in tumor cells that is common to different types of cancer, consisting of a characteristic combination of genes. According to the team’s paper in the journal Genome Biology, the algorithm also detected types of genes in the pattern that had never been clearly linked to cancer before.
The team ultimately used data from lung and colorectal cancer cells to train the algorithm before applying it to data sets of other kinds of tumors.
The project aims to go far beyond the classification of “healthy” versus “cancerous” cells. In initial tests, ikarus already demonstrated that the method can also distinguish other types (and certain subtypes) of cells from tumor cells. “We want to make the approach more comprehensive,” Akalin says, “developing it further so that it can distinguish between all possible cell types in a biopsy.”
In hospitals, pathologists tend only to examine tissue samples of tumors under the microscope in order to identify the various cell types. It is laborious, time-consuming work. With ikarus, this step could one day become a fully automated process. Furthermore, Akalin notes, the data could be used to draw conclusions about the tumor’s immediate environment. And that could help doctors to choose the best therapy.
- Meet ‘VALHALLA’, a Machine Learning Method That can Hallucinate an Image of Written Words, and Then Use It to Help Translate The Text into Another Language
- Tanushree Shenwai
- June 10 2022
Notes:
Machine translation is a branch of computational linguistics that uses software to convert text or speech between languages.
Typically, machine translation replaces words in one language with words in another. However, this method rarely results in a decent translation because recognition of entire phrases and their closest counterparts in the target language is required. Many words have several meanings, and not all terms in one language have comparable words in another.
Many researchers have been working to solve this challenge using corpus statistical and neural techniques, which has led to better translations, linguistic typology handling, idiom translation, and anomaly isolation. Typically, these methods require source phrases to be linked with corresponding images during training and testing. This specifically limits their utility in situations when images are not available during inference.
In their recent work, the researchers first explore whether a system that only has access to images during training time can generalize to these situations in their latest work. “Visual hallucination, or the ability to conceive visual scenes, can be used to improve machine translation systems,” they claim. Further, they state that if a translation system had access to images during training, it could be taught to abstract an image or visual representation of the text sentence to ground the translation process. This abstracted visual representation could be utilized instead of an actual image to perform multimodal translation during the testing period.
The researchers present a basic but effective VisuAL HALLucinAtion (VALHALLA) framework, which is based on machine learning for machine translation that integrates visuals during training to build a more successful text-only model. In machine translation, the models are trained to augment the text representation recovered from the source phrase with a latent visual representation that is similar to the one extracted by an multimodal translation system from a real image.
Their findings show that discrete visual representations work better than continuous visual embeddings currently used in multimodal translation approaches. They demonstrated the superiority of VALHALLA over strong translation baselines on three typical machine trainslation datasets with a wide variety of language pairs and different training data sizes.
Additional research indicates that, in limited textual contexts, VALHALLA models indeed use visual hallucination to improve translations.
- How AI Is Useful — and Not Useful — for Cybersecurity
https://www.darkreading.com/attacks-breaches/how-ai-is-useful-and-not-useful-for-cybersecurity
- Howie Xu
- June 9, 2022
Notes:
AI was introduced to the center stage of the cybersecurity industry a few years ago, originally to tackle malware detection and anomaly detection use cases. We have come a long way to better understand both the usefulness and the limitations of applying AI to cybersecurity, especially in the zero-trust era.
First, a zero-trust architecture doesn’t remove the need for AI. Though zero trust eliminates the attack surface and reduces the chance for the anomaly to happen, zero trust demands AI more.
In the zero-trust era, we need a personalized, contextual, dynamic, and granular security policy — which is exactly what zero trust is about. Access control, for instance, is no longer based on simple rules but a set of complex policies based on your identity, your device, your posture, your intention, your risks, your content, and a lot of rich data points.
However, generating such complex, granular, and personalized policy at scale can be very time-consuming if relying on human rules and heuristics. Different employees will use different applications and such application usage may need to evolve fast in a short period of time. AI is a critical technology to make such an intelligent and personalized security policy recommendation at scale.
At the same time, it is impossible for AI to capture or comprehend all the nuances and contexts of any complex environment, so AI may make recommendations that are suboptimal from experts’ eyes. With ongoing human feedback, we can improve the AI model and its effectiveness.
zero trust gives the enterprise much tighter protection than it has had in the past, but no matter how tightened things are, there is always a weak link somewhere. Therefore, we want AI to assist with evasive and unknown threat detection and prevention.
Enterprise customers want to utilize AI in a way that is easily understood and digestible by security professionals. The “explainable AI” may not improve the AI model efficacy on the surface, but it will increase the adoption of AI significantly.
AI is useful to scale the enterprise security functions, like more-intelligent policies and more-intelligent threat detection as discussed above. AI works best when security professionals and AI are complementing each other. In the end, AI is an assistant to security professionals and will not be a replacement for human effort for a long time to come.
Notes:
Chief human resources officers (CHROs) and the organizations they lead are looking to build the expertise they need by upskilling talent. Add to those challenges getting internal mobility right, providing employees with learning and growth opportunities, coaching managers to be talent champions, achieving less bias in hiring decisions and the future of work’s growing challenges become clear.
Finding new ways to improve upskilling and internal mobility, remove biases in the hiring and retention process and provide employees with a roadmap of what’s next in their careers is core to the future of work
The potential AI has to remove biases and give women worldwide the opportunity to make the most of their talents and change the direction of their and their families’ lives is encouraging. In addition, leading enterprise and search companies use the Talent Intelligence platform to find new candidates with the specific capabilities, skills and strengths needed to excel in highly technical roles.
Masking for bias, the leading candidates are often women with Masters and Ph.D. degrees in AI, computer science, machine learning and mathematics attending universities worldwide. Getting hired for a senior technical role in an enterprise software company changes the growth trajectory of their careers, elevating an entire family economically at the same time.
Biases start in the data sets used for hiring decisions yet can also be impacted by the conscious and unconscious biases of hiring managers and HR professionals. AI-based hiring algorithms need data sets larger than a single company can provide, as Amazon learned. HR and talent management professionals at Cultivate ’22 agreed that AI needs to be trained to focus on creating stronger connections between career paths and skills. That approach increases an applicant’s potential for success in their role. While skills-focused AI won’t eradicate conscious and unconscious biases from hiring decisions, it’s the most promising direction, as it’s using the technology to predict where a candidate with a specific skill will excel and which roles are the next best ones for them.
Helping employees identify their innate capabilities and skills, then providing them with personalized skill plans, is core to the future of work and talent management. Employees know their capabilities and skills define their careers, not their current job position or the company they work for. An employee’s ability and willingness to learn and re-learn define the future of work today. They’re looking for employers who will invest in their development and give them opportunities to excel and earn more while progressing their careers.
The future of work is now balanced in favor of independent, always-learning employees who can define their career path based on their capabilities and skills. They’re no longer dependent on a company for their career. For talent management and HR professionals, this presents a daunting challenge. External rewards, including larger offices, more perks in an office and more pay, don’t matter as much as personal growth and autonomy. It’s here where AI and talent intelligence platforms are making a difference. Talent management and HR professionals rely on talent intelligence platforms for a wide variety of tasks across the spectrum of talent management.
