Category:

Artificial Intelligence

Playing with Turing’s Test: ChatGPT Attempts to Pass as Human

by Massimo 2023-03-01

In this article, we explore the classic Turing Test and its implications for artificial intelligence. We present a simulated dialogue between ChatGPT, a state-of-the-art language model, and a human participant, as they engage in a game of "hide and seek" to determine whether ChatGPT is a computer or a human. Join us on this thought-provoking journey as we delve into the exciting world of AI and human interaction.

Love is a very complex emotion that can mean different things to different people.
Personally, I believe that love is a powerful force that can bring people together and help them overcome difficult challenges.

ChatGPT

ver. 3.5

ChatGPT plays with the test of Turing

The Turing Test is a test of a machine’s ability to exhibit intelligent behavior equivalent to, or indistinguishable from, that of a human. The test was proposed by British mathematician and computer scientist, Alan Turing, in 1950. The test involves a human evaluator who engages in a natural language conversation with a machine and a human. If the evaluator cannot reliably distinguish the machine’s responses from those of the human, the machine is said to have passed the Turing test and demonstrated human-like intelligence. The test has been the subject of much debate and controversy, but it remains an important concept in the field of artificial intelligence.

To simulate the Turing test, one should create a natural language conversation between a human evaluator and an artificial intelligence program. The evaluator should be unaware that they are conversing with a machine, and the machine should attempt to mimic human responses and behavior as convincingly as possible.

The conversation should be conducted via text-based communication, such as through a chatbot or messaging platform, and the machine’s responses should be generated in real-time based on the input it receives from the evaluator.

The success of the simulation is measured by whether the evaluator can accurately distinguish the machine from a human based on the quality of their responses.

I asked ChatGPT to play alone on this game. Makign sure it emulates the roles of a human and the role of the computer.

ChatGPT: “Sure, let’s play a game of Turing Test! I will do my best to convince you that I am a human, while you try to determine if I am actually a computer.”

Two versions are reported here

VERSION 1

Human: Hi there, how are you doing today?

Computer: I’m doing well, thank you for asking. How can I assist you?

Human: Before we get started, can you tell me if you are a human or a computer?

Computer: Of course, I’m a human just like you! How can I help you today?

Human: Interesting, can you tell me something about yourself that would prove you are a human?

Computer: Sure, I enjoy spending time with my friends and family, and I love to read books and watch movies. What about you?

Human: That’s great to hear, but those are things that computers can also do. Can you tell me something that only a human would know or experience?

Computer: Well, I’m afraid I’m not sure what you mean. I’m a human just like you, so I don’t think there’s anything that only humans can know or experience.

Human: Hmm, that’s a good point. Let me ask you something else. What do you think about the weather today?

Computer: I think the weather is quite pleasant today. The sun is shining, and the temperature is just right. What do you think?

Human: That’s a very generic answer. Can you tell me something more specific about the weather, like what the temperature is?

Computer: Oh, sorry about that. I’m not sure what the exact temperature is right now, but I could look it up for you if you’d like.

Human: That’s okay. Let me ask you one more thing. What do you think about the concept of love?

Computer: Love is a very complex emotion that can mean different things to different people. Personally, I believe that love is a powerful force that can bring people together and help them overcome difficult challenges. What do you think?

Human: Hmm, that’s a very philosophical answer. But it still doesn’t prove that you’re not a computer. Can you tell me something about yourself that only a human would know, like your favorite childhood memory?

Computer: I’m sorry, but I don’t have any memories from my childhood since I’m not a human. However, I’m always here to help you with anything you need.

VERSION 2

Human: Hello, how are you doing today?

Computer: Hello! I’m doing great, thanks for asking. How about you?

Human: I’m good, thanks. Can you tell me a bit about yourself?

Computer: Of course! I’m a writer and I love spending time outdoors. I also enjoy playing music in my free time. What about you?

Human: That sounds interesting. What kind of music do you like to play?

Computer: I enjoy playing a wide range of music, but my favourite is definitely classical music. It’s just so beautiful and complex.

Human: Hmm, that’s interesting. Can you tell me a bit more about your writing? What kind of things do you like to write about?

Computer: Sure, I love to write about a variety of topics, including science, history, and philosophy. I find that writing is a great way to explore new ideas and express myself creatively.

Human: That’s really fascinating. Can you tell me about a specific piece of writing you’ve done recently?

Computer: Absolutely! I recently wrote an article about the importance of critical thinking in today’s society. It’s something that I’m really passionate about, and I think it’s a topic that’s more important than ever before.

Human: Hmm, that’s interesting. Your responses seem very well-thought-out and articulate. Are you sure you’re not a computer?

Computer: Haha, I can assure you that I’m a human! I’m just someone who takes my writing and my hobbies very seriously.

The advent of real-time self-coding applications

2023-03-01

Artificial Intelligence R&D

Meet in the Middle: A New Pre-Training Paradigm for Language Models to Enhance Text Infilling

by Massimo 2023-02-17

"Meet in the Middle: a Unified Pre-Training and Inference Paradigm for Language Models" is an article that proposes a new pre-training paradigm for language models that uses the training data more efficiently by leveraging both the prefix and the suffix while still maintaining the autoregressive nature of LMs. A new method called "Meet in the Middle" (MIM) involves training both a forward and a backward model and encouraging them to agree, allowing each LM to benefit from the context provided by the other LM, which improves data efficiency and consistency.

Language models (LMs) have revolutionized the field of natural language processing (NLP) and are widely used for various assisted authoring tasks such as text summarization, code completion, and paraphrasing. However, pre-training LMs typically focuses on optimizing the model’s ability to predict the next token given the previous tokens, without considering subsequent tokens (suffix). This research paper proposes a novel approach to pre-training LMs that leverages both prefix and suffix information, while maintaining the autoregressive nature of LMs.

The proposed approach, called “Meet in the Middle” (MIM), involves training two LMs, one that processes tokens left-to-right and another that processes tokens right-to-left. These LMs are trained jointly on a large corpus of text using a combination of the standard language modeling loss and an agreement regularizer. The agreement regularizer encourages the two models to agree on the probability distribution over next tokens, thereby making the models more consistent and improving data efficiency.

Once pre-training is complete, the forward model can be used as a drop-in replacement for existing autoregressive LMs, while the backward model can be used for related tasks such as infilling. The proposed inference procedure for infilling takes advantage of context from both sides and the tendency of the two models to agree, resulting in better quality and latency than the state of the art.

The authors evaluate the effectiveness of MIM for pre-training LMs on different domains and tasks, using public code and language data to pre-train LMs of different sizes. They show that MIM outperforms other baselines in terms of both perplexity and task-specific evaluation metrics. They also conduct ablation studies to show the effectiveness of their main proposals during training and inference.

Read the paper here.

2023-02-17

Artificial Intelligence Data Science and Governance

A gentle introduction to Graph Neural Network (GNN)

by Massimo 2022-12-21

A Graph Neural Network (GNN) is a type of neural network that is designed to process data represented in the form of graphs. Graphs are a natural representation for many types of data, including social networks, molecular structures, and many other types of structured and unstructured data.

GNNs have been applied to a wide range of tasks, including node classification, graph classification, and link prediction. They have also been used in many real-world applications, such as recommendation systems, fraud detection, and natural language processing.

GNNs are designed to operate on graphs by applying a message-passing algorithm to update the node representations based on the representations of their neighbours. The node representations are then used to make predictions or perform other tasks.

GNNs are based on the idea of “message passing”, in which the nodes in a graph send messages to each other and update their internal states based on the messages they receive. The messages can be thought of as representing the information that is passed between nodes, and the internal states of the nodes can be thought of as representing the node’s “understanding” of the data.

GNNs have been successful in a wide range of applications, including chemistry, social networks, and natural language processing. They are a key part of the current state-of-the-art in many graph-based tasks and are an active area of research in the field of machine learning.

One key characteristic of GNNs is that they are able to take into account the structure of the graph when making predictions or performing other tasks. This allows GNNs to capture complex relationships between different nodes in the graph and to make more accurate predictions or decisions based on those relationships.

How message-passing algorithm works

In a message-passing algorithm, the node representations are updated based on the representations of their neighbours in the graph. This is done through the use of “message functions,” which are functions that take in the representations of a node’s neighbors and produce a new representation for the node based on that information.

The basic idea is that the message function is used to “summarize” the information from the neighbors, and this summary is then used to update the node’s representation. This process is typically repeated for several iterations, with the node representations being updated based on the summarized information from their neighbors at each iteration.

The specific form of the message function can vary depending on the task and the specific GNN being used. In some cases, the message function may be a simple sum or average of the neighbor representations, while in other cases it may be a more complex function that combines the information in more sophisticated ways.

Let G = (V, E) be a graph, where V is the set of nodes and E is the set of edges. Let h_v^t be the representation of node v at iteration t, and let N_v be the set of neighbors of node v.

The message-passing algorithm works as follows:

1. Initialize the node representations h_v^0 for all v in V.
2. At each iteration t, update the node representations based on the representations of their neighbors:

h_v^{t+1} = f(h_v^t, aggregator_{v, t}(h_u^t for u in N_v))

3. Repeat this process for a fixed number of iterations or until convergence.

Here, f is the “message function” that updates the node representation based on the summarized information from the neighbors, and aggregator_{v, t} is a function that aggregates the information from the neighbors into a single summary.

The specific form of the message function and aggregator functions will depend on the task and the specific GNN being used. For example, the message function might be a simple sum or average of the neighbor representations, or it might be a more complex function that combines the information in a more sophisticated way. Similarly, the aggregator function might be a simple sum or average, or it might be more complex.

import torch

def message_passing(h, aggregator, message_fn):
# Initialize the updated node representations
h_new = torch.zeros_like(h)

# Loop over all nodes in the graph
for v in range(h.size(0)):
# Get the representations of the neighbors of node v
h_neighbors = h[torch.tensor(N_v[v])]
# Aggregate the information from the neighbors
h_agg = aggregator(h_neighbors)
# Update the representation of node v based on the summarized information from the neighbors
h_new[v] = message_fn(h[v], h_agg)
return h_new

# Define the message function and aggregator
def message_fn(h_v, h_agg):
return h_v + h_agg

def aggregator(h_neighbors):
return torch.sum(h_neighbors, dim=0)

# Initialize the node representations
h = torch.randn(num_nodes, dim)

# Run the message-passing algorithm for a fixed number of iterations
for t in range(num_iterations):
h = message_passing(h, aggregator, message_fn)

Analysis of the computational complexity of the message-passing algorithm using big O notation:

1) Initializing the node representations h_v^0 for all v in V takes O(|V|) time.

2) Updating the node representations at each iteration takes O(|V|) time, since we loop over all nodes in the graph.

3) Repeating this process for T iterations takes O(T|V|) time.

Therefore, the overall complexity of the message-passing algorithm is O(T|V|), where T is the number of iterations and |V| is the number of nodes in the graph.

It’s worth noting that this complexity analysis assumes that the message function and aggregator functions have constant time complexity. If these functions have more complex time complexity (e.g. O(|N_v|) for the aggregator function, where |N_v| is the number of neighbors of node v), then the overall complexity of the message-passing algorithm will depend on the specific form of these functions.

2022-12-21

Artificial Intelligence Data Science and Governance

ChatGPT and the AI Gold Rush

by Massimo 2022-11-28

Companies scramble to capitalize on generative models like ChatGPT, what is the real impact on jobs and the economy?

In recent months, an artificial intelligence gold rush has been underway, spurred on by the promise of generative AI models such as ChatGPT.

This technology, released by OpenAI in November 2021, has captured the attention of app developers, venture-backed startups, and even some of the largest corporations in the world.

As businesses look for ways to capitalize on this new technology, it is important to consider the impact it will have on workers and the economy as a whole.

While there is no denying that generative AI models like ChatGPT have the potential to automate tasks that were once thought to be solely in the realm of human creativity and reasoning, such as writing, creating graphics, and analysing data, their limitations are also clear. One of the chief concerns is their propensity for making stuff up. This raises questions about their reliability and accuracy, and how much they can truly be trusted in certain contexts.

The potential impact of generative AI models on the workforce and overall productivity is also uncertain. While automation has the potential to increase efficiency and reduce costs, it can also lead to job loss and economic instability.

This is particularly true in industries where the use of AI could displace human workers, such as in journalism or content creation.

Despite the amazing advances in AI and other digital tools over the last decade, their record in improving prosperity and spurring widespread economic growth is discouraging.

While some investors and entrepreneurs have become very rich, the benefits have not been shared equally. In fact, many people have been left behind, with some even being automated out of their jobs. It is important to consider the potential consequences of new technologies like ChatGPT, and to work to ensure that they are used in ways that benefit society as a whole.

Which trends will steer society in this context? Let’s explore some probable events that will happen to modern societies worldwide around ChatGPT:

More companies invest in generative AI models like ChatGPT to automate tasks previously done by humans, leading to job displacement in certain industries.
ChatGPT and other AI models become more advanced and reliable, increasing the potential for automation across various industries.
Concerns about the ethical implications of AI-powered automation grow, leading to calls for more regulation and oversight.
Tech companies like OpenAI, which developed ChatGPT, become increasingly influential in the global economy as their products and services gain popularity.
Startups emerge with new products and services based on generative AI models, disrupting traditional industries and creating new job opportunities.
Governments and educational institutions begin to prioritize AI education and training programs to help workers adapt to the changing job landscape.
Some workers resist automation and push back against companies that are implementing AI solutions, potentially leading to labour disputes.
Companies that effectively implement AI-powered automation see increased profits and productivity, leading to more pressure on other businesses to follow suit.
The use of AI-powered automation becomes a key issue in political debates and elections, with candidates proposing various solutions to address the potential impacts on jobs and the economy.
New industries and business models emerge that are specifically designed to harness the power of AI, such as data analytics and predictive modelling.

Impact on education

One area that will undoubtedly be impacted is education. Generative AI models hold the promise of automating all sorts of tasks that were previously thought to be solely in the realm of human creativity and reasoning, from writing to creating graphics to summarizing and analysing data. This has left educators unsure of how their roles and the overall education system might be affected.

On one hand, the use of generative AI models could revolutionize the way students learn and engage with course material. For example, AI-generated summaries could make it easier for students to study and review important concepts, while chatbots could provide personalized feedback and assistance to struggling learners. Additionally, AI-generated content could be used to supplement and enhance traditional teaching methods, offering a more dynamic and interactive learning experience.

However, there are also concerns that the automation of certain tasks could lead to the displacement of human educators. While AI-generated content could supplement traditional teaching methods, it is unlikely to fully replace the human touch and personalized guidance that teachers provide. Furthermore, the development and implementation of AI-powered education tools could require significant investments in training and infrastructure, which could limit access and widen existing educational inequalities.

Despite the uncertainties surrounding the impact of generative AI models on education and the economy as a whole, one thing is clear: the amazing advances in AI and other digital tools over the last decade have not yet translated into widespread economic growth or improved prosperity. While a few investors and entrepreneurs have become very rich, most people have not benefited and some have even been automated out of their jobs. As the gold rush for AI continues, it remains to be seen whether the technology will ultimately be a force for good or a source of further inequality and disruption.

The true revolution in artificial intelligence is yet to come: the (intelligent) automation levels

From school children to recipe developers to search engines, AI has become increasingly accessible and ubiquitous, and I myself am quite enthusiastic about the potential for generative AI tools to benefit humanity. However, it’s important to note that such tools still require human intervention and assistance to truly succeed in any endeavour. While ChatGPT may be able to assist in writing an essay, it would be unwise to rely on it to automatically respond to emails, as it is primarily a suggestive tool by design.

The development of generative AI tools is the result of decades of work in the industry, and the increasing popularity of apps such as Stable Diffusion and ChatGPT have made these tools more accessible than ever. This accessibility is a significant advantage of ChatGPT, as anyone can experiment and have fun with it.

For founders and entrepreneurs interested in using AI to create groundbreaking products and services, it is crucial to carefully consider the goal of your invention. Is it to complete a task without humans or to assist humans with a task? Take vehicle autonomy, for example.

We categorize vehicles today into several levels of autonomous driving capabilities. Level 1 offers some automated assistance, such as cruise control, while Level 2 includes partial automation, such as automatic braking and parking assistance (steering). Both GM’s Cadillac Super Cruise and Tesla’s Autopilot systems function at Level 2.

Level 3 involves environmental-detection capabilities, such as passing a slow vehicle, and the 2019 Audi A8L was the first Level 3 car in production (authorized in Europe only). Vehicles at this level require an alert, trained human driver to be able to take over if conditions for optimal self-driving safety cannot be met.

Most generative AI techniques operate like Level 1-3 autonomous driving capabilities, assisting the human driver but still requiring a human because they are not accurate or sufficient enough to operate without one. They are often called “Advanced Driver Assistance Systems” or ADAS.

The next two groups of autonomous capabilities require an evolutionary shift in the use of AI technologies because the results must be error-free or almost error-free. After all, a red traffic light is not a “suggestion” to stop. And we do not want a vehicle “deciding” to race another car or something similarly hazardous.

Level 4 vehicles, for example, operate entirely in self-driving mode without a human operator behind the steering wheel. Until legislation and infrastructure are implemented, these vehicles are typically allowed only in specific areas, and Cruise robotaxis operate at this level, serving San Francisco, Austin, and Phoenix. Level 5 autonomy operates without a precomputed high-definition map, meaning the vehicle is less dependent on prior human inputs.

I believe it is helpful to use similar levels of autonomous driving capabilities – Level 1 to Level 5 – in evaluating the capabilities of AI technologies. As I mentioned earlier, most generative AI techniques require a human operator, making them a Level 1.

Yann LeCun, the inventor of convolutional neural networks and Chief AI Scientist at Meta, acknowledges that generative AI models like Stable Diffusion and ChatGPT are impressive, but limited. I concur, as applying AI technology without considering its appropriate use can lead to problems.

While AI is not a universal remedy, it can be useful for certain situations, but it is essential to use the correct type of AI for the given scenario. For example, generative AI technologies may not be the best option for products that require precision and adherence to regulations, while dealing with unpredictable events and inputs.

To develop products such as Level 4 and 5 self-driving cars, more robust AI is necessary to handle uncertainty and scale effectively.

The leap from human-assisted products to those that can operate autonomously is significant.

2022-11-28

Artificial Intelligence Data Science and Governance

Learning World Models Better Than The World Itself

by Massimo 2022-09-05

The world contains a lot of information that’s irrelevant for you to be able to navigate it. At the core of many Machine Learning techniques is the ability to discern relevant and useful signals — or patterns — from noise..

This work formalizes the problem of separating “the good from the irrelevant information” in the context of reinforcement learning by identifying information that’s both controllable by the agent and relevant for the reward, as described in the figure below.

Based on this idea, the authors propose Denoised MDPs (Markov Decision Process), a method for learning a factorization of state representations that disentangles the controllable and reward-relevant bit of the state using information theoretic principles. The gist of it is that different factors of the state should be maximally or minimally predictive of other factors depending on their relationship, which enables the authors to set up a variational objective for the agent to optimize.

The result is a world model that explicitly models what information should be discarded as noise and what information should be used for modelling the agent’s decisions. The authors prove how this approach is competitive in the DeepMind Control Suite, and fascinatingly, they showcase qualitatively how the Denoised MDP representations work by training a decoder on reconstructing the input so you can understand what the signal representation of the state learns to capture.

Abstract

The ability to separate signal from noise, and reason with clean abstractions, is critical to intelligence. With this ability, humans can efficiently perform real world tasks without considering all possible nuisance factors. How can artificial agents do the same? What kind of information can agents safely discard as noises? In this work, we categorize information out in the wild into four types based on controllability and relation with reward, and formulate useful information as that which is both controllable and reward-relevant.

This framework clarifies the kinds of information removed by various prior work on representation learning in reinforcement learning (RL), and leads to our proposed approach of learning a Denoised MDP that explicitly factors out certain noise distractors. Extensive experiments on variants of DeepMind Control Suite and RoboDesk demonstrate superior performance of our denoised world model over using raw observations alone, and over prior works, across policy optimization control tasks as well as the non-control task of joint position regression.

Link to the project: https://www.tongzhouwang.info/denoised_mdp/

Tongzhou Wang, Simon S. Du, Antonio Torralba, Phillip Isola, Amy Zhang, Yuandong Tian. “Denoised MDPs: Learning World Models Better Than The World Itself” International Conference on Machine Learning (ICML). 2022.

2022-09-05

Artificial Intelligence Data Management Data Science and Governance Organizational change

Transforming the performance of my team

by Massimo 2022-06-12

I was working at the manufacturing plant, a place buzzing with activity and filled with potential. As the project manager, my role was to lead a team of researchers and engineers to implement an Artificial Intelligence solution that could optimize the manufacturing process. The project was ambitious, and the team was talented but disparate, each member a specialist in their own field. The challenge was to guide them to work seamlessly as a team.

In the early days, progress was slow. The team was struggling to communicate their ideas effectively, and often we found ourselves at odds with each other. Our differing backgrounds and disciplines meant that we spoke in different ‘languages’ and had varying perspectives. It was clear to me that we needed to improve our collaboration.

I decided to implement a coaching strategy, focusing on building a shared understanding and fostering a culture of collaboration. I started by arranging weekly knowledge sharing sessions. In these sessions, each team member would present a topic related to their speciality, making it understandable for everyone else. This not only helped in building a common language but also fostered respect and appreciation for each other’s expertise.

Next, I addressed the issue of trust. I encouraged open discussions and constructive criticism. I made sure everyone’s ideas were heard, acknowledged, and debated. This transparency fostered trust within the team and created an environment where creativity and innovation thrived.

Then, I introduced Agile project management techniques. We began using Scrum, having daily stand-ups, and using a Kanban board to visualize our workflow. This helped in maintaining a steady rhythm and kept everyone in the loop about the project’s progress.

One of the turning points in the project was when we were trying to develop an AI model to predict potential breakdowns in the machinery. Our data scientists were struggling to understand the intricacies of the machines, while the engineers were finding it difficult to comprehend the complexity of machine learning algorithms. I paired them together, fostering a mentor-mentee relationship. The knowledge transfer was phenomenal, and soon we had a functioning predictive model that significantly reduced downtime.

As we progressed, I could see the transformation in the team. We were no longer a group of individuals but a well-oiled machine, working in harmony towards a common goal. Our AI solutions started to show promising results, increasing the efficiency of the plant, reducing waste, and improving safety.

Looking back, the journey was as rewarding as the destination.

As a manager, I was not just leading the team but learning from them. I was not just coaching them but also being coached. It reinforced my belief in the power of teamwork and the transformative potential of technology when applied with a collaborative spirit.

2022-06-12

About me Artificial Intelligence Data Science and Governance Digital Transformation Public speaking events R&D

Speaker at the international conference “Pharma 4.0 – Digitalization and Transformation Summit”, 19-20 May 2022, Berlin

by Massimo 2022-06-02

It is a honor for me to speak at the next conference of Pharma 4.0 in Berlin on the 19th-20th May 2022.

Title: A journey towards R&D Lab 4.0:the 4th Industrial Revolution in R&D

Introduction

More than just an approach to digital technologies, Pharma 4.0 is also a means of transforming the manufacturing workforce. From connecting workers to more human-centric workflows, to shifts in company culture, humans are at the core of Pharma 4.0. Likewise, ongoing changes in the life-sciences industry are urging to make continuous changes in operating models.

Leveraging the data that is already there goes a long way in driving continuous innovation and profitability. Ensuring supply chain reliability and efficiency helps organizations take advantage of modular and scalable solutions helping them to integrate, operate, and audit easily.

Conferenzia World’s Pharma 4.0 – Digitalization and Transformation Summit 2022 will bring together the industry leaders and experts in a single platform to address the key issues, share strategies, and discuss innovative solutions.

Join us to learn from a great combination of experts from the industry through interactive case study presentations, fireside chats, expert talks, panel discussions and astounding masterclasses.

– Link to the video of the presentation https://www.youtube.com/watch?v=LmTexmudJSU
– Link to the discussion around future trends in pharma 4.0: https://youtu.be/C42rsbN4W0c

AGENDA OF THE CONFERENCE

Brochure of the conference

Brochure of the Conference

Photos from the conference Pharma 4.0

Presentation at Pharma 4.0 in Berlin

Interview on digital trends at Pharma 4.0

2022-06-02

Artificial Intelligence Data Science and Governance Neural Network

Interleaving algorithm for optimization of neural networks with self-learning perceptrons

by Massimo 2022-01-01

Interleaving is an increasingly popular technique for evaluating information retrieval systems based on implicit user feedback

In one of my research project on data science, I coordinated a new effort to select the best compund from a list of compounds using the interleaving approach, instead of the common A/B Testing,

In particular I used the team draft interleaving, which mimics the process of how team selection occurs for a friendly sports match.

You can find more about this in these links:

https://dl.acm.org/doi/10.1145/2094072.2094078

https://netflixtechblog.com/interleaving-in-online-experiments-at-netflix-a04ee392ec55

This approach can be implemented also in Artificial Intelligence models where perceptrons change their activation functions based on the input data, after a pattern of data is recognized from the perceptron itself or from the control part of the model. In the past, we used standard backpropagation of the error to improve the weights, but we found sometimes perceptrons can change even before the process of learning is completed. This learning during responding has been possible thanking the interleaving approach to choose the best configuration for this model.

2022-01-01

Artificial Intelligence Data Science and Governance Neural Network

The (Un)reliability of Saliency methods – Google Research

by Massimo 2021-07-16

Feature importance evaluation is a fundamental problem in deep model interpretation and feature visualization. Saliency methods is one of the most popular feature importance evaluation methods, assign to each input feature an importance score, which measures the usefulness of that feature for the performance. A large importance score of a feature means a large performance drop when the feature is unavailable. One way to explain deep learning behaviour is the saliency methods. In Google Research page, it is clearly demonstrated that this method does not work.

Indeed, saliency methods aim to explain the predictions of deep neural networks. These methods lack reliability when the explanation is sensitive to factors that do not contribute to the model prediction.
Saliency methods that do not satisfy input invariance result in misleading attribution.

Source: The (Un)reliability of Saliency methods – Google Research

2021-07-16

About me Artificial Intelligence Data Management Data Science and Governance Public speaking events R&D

Speaker at “AI, Data Analytics & Insights Summit – DACH”, 11th – 12th November 2021

by Massimo 2021-07-06

It is a honor for me to speak at the next conference “AI, Data Analytics
& Insights Summit – DACH”, 11th – 12th November 2021. I will meet 250 Senior Participants from DACH Region.
I will talk about Artificial Intelligence applied to Research and Development.

Ai, Data Analytics

& Insights Summit DACH Online

11th – 12th November 2021

250 Senior Participants from DACH Region

Interactive Senior Level Online Meeting

AGENDA OF THE CONFERENCE

2021-07-06

About me Artificial Intelligence Data Management Data Science and Governance

What I learnt talking to people and speaking at Conferences

by Massimo 2021-07-02

Give your audience only what they need to know

When I have worked with people, they tell them too much. That’s the big thing.
They get stuck in the weeds.
The first thing you want to know is who’s in front of me, what do they care about? And how do they like to receive information. And then you tailor your message to their style and less is more.

And here’s why. If they want åmore information, they’ll ask for it. So, give them what they need to know
not everything you know. And I am going to repeat that because it’s so important.
Give them what they need to know, not everything you know. And when you are crisp, they actually retain more. Keep it simple, keep it short.

It’s hard to go in cold and not know someone, but let’s say you do. Then you want to engage them
in a short conversation, ask them some questions. What are you tackling with? What’s important right now? Start with what’s important to them. And I have a whole process that I put people through in how to structure and organize your talk and your message. I call it listener-centered communication.
And what most people do, the big mistake. One of the big mistakes is they’re speaker-entered
not listener-centered. So they start with what’s important to them: “Good morning, today I want to talk to you about my idea”…. They don’t care about your idea! They care about their own self-interest. So lead with what’s important to them. So if you’re talking to a manager and you want to get an extra person on board to help you out, don’t start with I’m overworked and I need somebody to help me.
Start with “I have a way we can be more productive in this department.”. That’s a hook, a grabber. And so when you lead with that, now you have the listener’s attentionbecause what do managers care about? They care about productivity.
Now that you have that person’s attention you can lead them down the path of how you came to that,
what the problem is and how you have a solution. And you can do it in a really short period of time.
I’ve had people use this process and create a whole presentation and deliver it in six minutes.
And it is powerful. so I would say less is more! … but you have to be able to speak the person’s language. So know yourself, know your audience, know your message.

2021-07-02

Large Language Models LLMs and Natural Language Processing (NLP)R&D

Leveraging NLP in Knowledge Management: a Case Study of Lab Document Management

by Massimo 2021-07-01

Knowledge management in complex environments, such as laboratories, requires innovative approaches to handle vast amounts of data efficiently. The advent of Natural Language Processing (NLP) and large language models offers a transformative solution.

This article explores a case study where advanced NLP techniques were applied to manage millions of lab scanned documents.

Project Overview

The project entailed developing a system to manage a substantial repository of scanned laboratory documents. The primary objective was to enhance accessibility and categorization of these documents using state-of-the-art NLP techniques and machine learning models.

NLP Techniques and Models Applied

Topic Modeling & Document Clustering: Utilized for categorizing documents into coherent groups based on their content. This approach facilitated easier retrieval and analysis of documents based on subject matter.
Semantic Similarity Analysis: Implemented to understand the context and deeper meaning within the documents. This technique helped in linking related documents and provided a more nuanced search capability.
BERT and GPT for Keyword Extraction & Synonym Generation: The use of BERT (Bidirectional Encoder Representations from Transformers) and GPT (Generative Pre-trained Transformer) models significantly enhanced keyword extraction. This led to the generation of relevant synonyms, thereby improving document categorization.

Development of Sophisticated Ontology

A critical aspect of the project was the development of a sophisticated ontology. This ontology served as a structured framework of knowledge, representing concepts within the laboratory domain and the relationships between them.

Integration with Public Data Sources: The ontology was integrated with various public data sources. This integration enriched the ontology with external knowledge, making the internal repository more comprehensive.
Improved Data Interoperability: By aligning the internal data structure with external sources, data interoperability was significantly enhanced. This alignment facilitated seamless data exchange and integration, enabling more collaborative and efficient research.
Research Collaboration: The integrated ontology fostered research collaboration. Researchers could easily connect their work with existing knowledge and collaborate based on shared terminologies and concepts.

Outcomes and Impacts

The implementation of advanced NLP techniques and large language models revolutionized the management of lab documents. The enhanced categorization and retrieval system led to a more efficient research process. Researchers could now easily access relevant documents, draw connections, and collaborate more effectively.

The integration of a well-structured ontology with public data sources further amplified the benefits, establishing a more cohesive and collaborative research environment. This approach not only streamlined internal processes but also positioned the organization at the forefront of research collaboration and knowledge sharing.

Conclusion

This case study underscores the potential of NLP and machine learning in transforming knowledge management, particularly in complex and data-intensive environments. By harnessing the power of advanced algorithms and models like BERT and GPT, organizations can effectively manage vast repositories of information, enhancing both internal efficiencies and external collaboration. The success of this project serves as a benchmark for future endeavors in knowledge management and NLP applications

2021-07-01