Collection of oral questions and exam material
Template
🇺🇸 2019-20
- Your questions here... (leave don't remove spaces for a correct visualization)
- Be the first to contribute, open a pull request with your oral questions or other relevant material!
🇺🇸 2018-19
Topic chosen by the student: Python (almost all, excluding GIL)
Questions on the assignments
- Given a bean, how can we tell what is its public API?
- Question about complexity of some haskell function. What are the problems between arrays and generics in Java? What happens with generic at runtime/after compilation?
- Is Python more OO or more functional, according to your opinion?
Questions on the syllabus
- Explain the concept (with written example) of covariance and contravariance in a language with universal polymorphism and explain in what cases their use is safe
- Explain inversion of control and dependecy injection
- What is lazy evalutaion in haskell and explain the spirit of IO Monads
Topic chosen by the student: lambda expressions in Java
Questions on Java
- What are streams in java?
- Example of lambdas in a context different from streams
- What are functional interfaces?
- How the java compiler manges lambdas?
- Differences between component, packages and classes
- Talk about the lifecycle of a sw component
- How can we interact with java beans?
- How can Netbeans (or another builder tool) provides "live interaction" with a bean?
- What kind of properties a bean can have?
Questions on Haskell
- What is "Functor" in haskell?
- What relationships there are between functor and maybe type class?
- Lazyness in haskell
Question on Python
- What are python decorator?
- How can we write functions with a variable number of arguments
Other questions
- Describe the different kind of parameter passing strategies
- What is memoization?
Topic chosen by the student: JVM internals and JVM instruction set.
For this exam, most of the questions were asked during the presentation of the topic chosen by the student.
Questions on Java
- Memory management
- How are lambda expressions implemented?
- Talk about streams
Question on Python
- What is GIL?
- Talk about decorators and higher order functions in general
- Namespaces and scopes
- What does the "@staticmethod" decorator do?
Other questions
- Differences between "reduce" in functional programming and "collect" in the Java Stream API
- Give an example of list comprehension and write a function using a functional language that does the same thing, using combinators such as filter, foldr, map ecc
- Be the first to contribute, open a pull request with your oral questions or other relevant material!
- Be the first to contribute, open a pull request with your oral questions or other relevant material!
- Be the first to contribute, open a pull request with your oral questions or other relevant material!
- Be the first to contribute, open a pull request with your material!
- Be the first to contribute, open a pull request with your oral questions or other relevant material!
- Be the first to contribute, open a pull request with your oral questions or other relevant material!
- Be the first to contribute, open a pull request with your oral questions or other relevant material!
- Be the first to contribute, open a pull request with your oral questions or other relevant material!
- Be the first to contribute, open a pull request with your oral questions or other relevant material!
🇺🇸 Before 2020
-
Differences between Apriori and FP-Grow. Hierarchical Transactional clustering -> Rock and differences with classic hierarchical clustering. MLE: expectation phase and maximization Phase, comparison with the same phases in KMeans
-
two questions about FP-Grow.
-
fp-growth; the difference between different kinds of clustering algorithms.
- Be the first to contribute, open a pull request with your oral questions or other relevant material!
- Be the first to contribute, open a pull request with your oral questions or other relevant material!
- Be the first to contribute, open a pull request with your oral questions or other relevant material!
- Be the first to contribute, open a pull request with your oral questions or other relevant material!
🇺🇸 2020-21
- Compression of WebGraph, meaning of "near pages", how to optimize the saving of the "extra nodes"
- Karp-Rabin fingerprints, properties, proof
- Pearson's Chi-Square test. Meaning of the tables used in this test
- Zsynch (exercise like in written exams, but explaining your thoughts while writing)
- Delta compression (zdelta)
- Delta compression for groups of files (the one with the graph)
- Be the first to contribute, open a pull request with your oral questions or other relevant material!
- Be the first to contribute, open a pull request with your oral questions or other relevant material!
- Be the first to contribute, open a pull request with your oral questions or other relevant material!
🇺🇸 2019-20
- Write the loss for a linear model in form of ridge regression (with Tikhonov regularization)
- Write (and briefly comment on) the most important factors ruling the flexibility of the SVM and of the NN approaches Compute dE/do_i for a NN (the same seen in class) showing the single steps of the derivation for the dE/do_i (assuming to have already delta for the units of the other layers indexed by h and k. Finally, write delta_k, delta_t, delta_i
- Define the VC-dim. Does the VC-dim in Phi_1 increase or decrease if the value of the regularization parameter increases? Explain.
- Write the net_t(x) of a preceptron with inputs i in [1, ..., k]
- Write the radial basis function kernel
- Answer true or false to the following and motivate your answer
- In a SVM can the alpha values help to select the best features? False (they only select some input patterns and not the components)
- To estimate the (future) predictive capability of your model is it a good practice to consider the result and accuracy obtained by the model selection phase without looking to the training results? False (only the test result cares to estimate the (future) predictive capability)
- Increasing the VC-dim, th VC-bound on the risk R (according to SLT) increases. True
-
Write the derivation of the bias-variance decomposition (assuming without proving the variance lemma)
-
Show a picture of:
- undercomplete autoencoder
- overcomplete autoencoder
-
Equation of the RBF Kernel. Equation of the Loss for Ridge regression + comments on the role of lambda.
-
Equations used for SOM. Why is the Neural Network able to approximate every function? Why do we use deep learning if one layer is enough?
- Be the first to contribute, open a pull request with your oral questions or other relevant material!
🇮🇹 La maggior parte dell'orale verte sul progetto e molte delle domande di teoria sono collegate.
🇮🇹 2019-20
-
Macro Data Flow con esempio
-
Buildin blocks per streming e data.
-
Point to point e collective comunication in fast flow. Qunado serve collective comunication.
-
Service time, latency, efficiency più esempi su quando voglio ottimizzare cosa
🇮🇹 2018-19
- Perchè non hai usato un manager e hai fatto fare tutto all’emitter? Con il manager hai dei movimenti di dati in più? L’emitter ha qualche guadagno se c’è il manager?
- Perchè parti subito con un numero fissato di worker? Non sarebbe stato meglio farlo lavorare uno e poi capire quanti worker servono dinamicamente?
- Sarebbe stato giusto fare un’analisi del tempo che ci metto per fare uno switch del numero di worker in modo da capire ogni quanto cambiare
- Se ho due map quali ottimizzazioni posso fare? Se faccio la map fusion poi ho un guadagno? Ce l’ho sempre?
- Se metto una map dentro ad una farm che guadagno ho se aumento il numero di worker della map o se aumento il numero di worker della farm? Se tolgo la map e lascio la farm aumentando i worker cambia qualcosa? L’efficiency come cambia? Se tolgo tutto e lascio solo un nodo sequenziale cambia qualcosa? L’efficiency come varia (diventa 1)?
- Differenza tra work span model e amdahl law, voleva sapere che i nodi devono avere lo stesso tempo di esecuzione
- Come si calcola lo speedup?
- Amdahl Law (con limite)
- Cosa devo aggiungere ad un farm che tenga in conto sia il tempo che il consumo energetico (considerando che entrambi dipendono dal nw)? Devo tenere conto di grado di parallelismo e frequenza del processore. Nel collector posso calcolare il service time Ts. Posso variare il numero di worker e vedere all'aumentare del nw come varia l'energia. Più worker = meno tempo di uso di CPU con più consumo. Posso fermarmi quando diventa sconveniente aumentare il nw per l'overhead.
- Tipi di pattern
- Vettorizzazione del codice
🇮🇹 2017-18
- Hamdal law
- Macro data flow, con un esempio di conversione fra "farm(pipe(f1,f2))", chiedendo di disegnare il grafico
🇺🇸 Most of the oral exam focuses on the project and many of the theory questions are related.
🇺🇸 2019-20
-
Macro Data Flow with examples
-
Buildin blocks for streming and data.
-
Point to point and collective comunication in fast flow. When to use collective comunication.
-
Service time, latency, efficiency and example where to use what metrics.
🇺🇸 2018-19
- Why didn't you use a manager and have the emitter do everything? Do you have more data movements with the manager? Does the emitter have any earnings if there is a manager?
- Why do you start immediately with a fixed number of workers? Wouldn't it have been better to have him work one and then figure out how many workers they need dynamically?
- It would have been right to do an analysis of the time it takes to make a switch of the number of workers in order to understand how often to change
- If I have two maps, which optimizations can I do? If I do map fusion then do I have a profit? Do I always have it?
- If I put a map inside a farm, what profit do I have if I increase the number of workers on the map or if I increase the number of workers in the farm? If I remove the map and leave the farm increasing the workers does something change? How does efficiency change? If I remove everything and leave only a sequential node, does something change? How does the efficiency vary (becomes 1)?
- Difference between work span model and amdahl law, he wanted to know that nodes must have the same execution time
- How is the speedup calculated?
- Amdahl Law (with limit)
- What should I add to a farm that takes into account both time and energy consumption (considering that both depend on the nw)? I have to take into account the degree of parallelism and the frequency of the processor. In the collector I can calculate the service time Ts. I can vary the number of workers and see how the energy changes as the nw increases. More worker = less CPU usage time with more consumption. I can stop when it becomes inconvenient to increase the nw for overhead.
- Types of patterns
- Vectorization of the code
🇺🇸 2017-18
- Hamdal law
- Macro data flow works, with an example of conversion between "farm(pipe(f1, f2))", asking to draw the graph
- Be the first to contribute, open a pull request with your oral questions or other relevant material!
- Be the first to contribute, open a pull request with your oral questions or other relevant material!
- Architectures for robot behavior described by the three primitive functions
- Motion control with gravity compensation (robot control)
- Potential fields (robot navigation)
- Histograms of an image (robot vision)
- ZMP components (bipedal locomotion)
- Be the first to contribute, open a pull request with your oral questions or other relevant material!
- Be the first to contribute, open a pull request with your oral questions or other relevant material!
- Be the first to contribute, open a pull request with your oral questions or other relevant material!
- Be the first to contribute, open a pull request with your oral questions or other relevant material!
- Be the first to contribute, open a pull request with your oral questions or other relevant material!
- Be the first to contribute, open a pull request with your oral questions or other relevant material!